On this day in 2006, AMD confirmed its acquisition of graphics chip firm ATI. AMD stumped up a cash and stock deal worth a total of $5.4B for the Canadian PC graphics innovators. With 20/20 vision now 20 years on, we can see the deal helped AMD prosper on three fronts: continuation and innovation of Radeon GPUs, the rise of the APU, and AMD’s dominance in the console business. That’s not all, of course, and it is also interesting to recall that AMD approached Nvidia before it bought ATI.

AMD CEO Hector Ruiz and ATI CEO Dave Orton appeared together in New York on the morning of July 24, 2006, to publicly announce the deal. Ruiz told the press that the deal, unanimously approved by the directors of both companies, would "reinvent our industry." The AMD CEO went on, “We believe AMD and ATI will drive growth and innovation for the entire industry, enabling our partners to create differentiated solutions and empowering our customers to choose what is best for them.” Orton added that “Joining with AMD will enable us to innovate aggressively on the PC platform.”

In the next couple of years, the ATI graphics brand slowly melted away. For example, the ATI R600 (Terascale 1, unified shader architecture) GPU, which was in development at the time of the acquisition, would become the Radeon HD 2900 series under AMD branding. As a transition product, some HD 2900 XT boxes would still carry ATI packaging and branding. Taking the cooling solution off an AMD graphics card, you might still see an ATI GPU under the thermal paste, all the way up to around 2010. In 2011, Tom’s Hardware published its 25 Years Of Graphics History: A Farewell To ATI, In Pictures, which is a must-read for fans of the era.

While graphics card development rolled on in a tit-for-tat battle with Nvidia, AMD’s next fortuitous chunk of synergy came from integrating its Radeon IP alongside its CPU cores to create APUs. It started this journey with the AMD Fusion line in 2011. Nowadays we have APUs that have taken this vision far further, with the Ryzen AI Max / Max+ (Strix Halo and Gorgon Halo) series. The integrated graphics on these processors pack up to 40 RDNA 3.5 Compute Units and can go toe-to-toe with desktop graphics like the RTX 4060/4070, depending on workload. They also benefit from unified memory, allowing users to configure oodles of VRAM, if they have it spare.

As the Radeon developers forged ahead moving from the GCN to RDNA graphics architecture era, AMD saw an opportunity in the console space. From the early to mid 2010s, AMD made inroads into APU development that meant the processors became attractive solutions for console developers. It still holds pretty tightly to that market today, which has spilled over to handhelds. However, Intel looks far more serious in this market now, with Panther Lake and B390 iGPUs. Moreover, Nvidia could surely make a dent on consoles with Arm plus GeForce semi-custom SoCs if it wasn’t living it up in the lucrative AI market.

Red and green would never be seen

Before the AMD and ATI deal was inked, reports indicated there were chances of a similar merger involving AMD and Nvidia. Our 2012 report on this ‘missed opportunity’ suggests Jensen Huang was being quite difficult during the negotiations. Apparently the man in the leather jacket insisted that he become chief executive of the combined company. That made Hector Ruiz pretty cool on the prospect, so AMD’s attention was diverted towards ATI.

The developers of RPCS3 have announced that their PlayStation 3 emulator’s minimum system requirements have been adjusted to include a crop of even older Radeon graphics cards. This well-regarded application already supported ATI Terascale 2 GPUs from the Radeon HD 5000 and 6000 series, but now gamers still cradling Radeon HD 2000, 3000, and 4000 series can join in the fun. Some of these GPUs date back nearly 20 years. Please note that while RPCS3 is available for a range of platforms, this extended ATI architecture support is only for Linux users.

Fine wine on Terascale 2

Back in 2021, the RPCS3 devs lowered the emulator’s system requirements to embrace Radeon HD 5000 and HD 6000 (Terascale 2) series graphics cards. In the subsequent five years, the emulator has been tuned to improve performance on hardware from these generations. An example provided asserts that when using Persona 5 as a benchmark, performance in Kamoshida’s Castle went from 35 to 49 FPS between 2021 and today (on a PC system with an ATI Radeon HD 5850, AMD FX-8350 CPU, and 8GB of RAM).

Inspired Terascale 1 experiments

Success spurred the devs to investigate TeraScale 1 performance and compatibility. Some OpenGL wrangling ensued due to TeraScale 1 GPUs only reporting OpenGL 3.3 and GLSL 3.30 support, but getting these old and unloved ATI cards to work with the RPCS3 OpenGL renderer was surprisingly simple. It was done by overriding the reported OpenGL levels to 4.3 and 4.40.

Initial tests made use of a system packing an ATI Radeon HD 4890 with 1GB of VRAM – the most powerful Terascale 1 GPU ever sold. According to the RPCS3 blog, this card has a more powerful GPU than the PS3, but the emulator’s complex shader code is not optimal for older architectures like this. Nevertheless, “Simpler games work perfectly fine, with playable performance,” notes the blog, “while games that have more complex graphics are GPU bottlenecked with subpar performance.” Screenshots show Metal Gear Solid HD Collection and Outrun Online Arcade with frame rates close to 60 FPS.

Moving down the stack and further back in time, GPUs like the ATI Radeon HD 3870 are “still able to render very simple games with playable performance in RPCS,” assert the devs. To achieve playability, you will want to dial in PSP-like resolutions, though. Specifically, 273p was mentioned.

What about the weakest TeraScale 1 GPUs? The ATI Radeon HD 2600 XT with a measly 256MB of VRAM even performs poorly at 273p. Screenshots from RPCS3 tests show it only managing 8 FPS in Outrun Online Arcade and 13 FPS in Portal.

With RPCS3’s progress over the last half decade, these performance numbers could also improve. In fact, the blog states that “future optimization should improve performance by quite a bit on these old GPUs.”

It looks like a good time to dust off your retro PC and try a few old PS3 classics on your ATI TeraScale 1 GPU-accelerated rig. Remember this extended graphics card support requires a Linux distro with the Mesa driver r600. Then you need to launch RCPS3 with the following environment variables: MESA_GL_VERSION_OVERRIDE=4.3 MESA_GLSL_VERSION_OVERRIDE=440 ./rpcs3.AppImage

Providing another fine example of the depth of support for older hardware on Linux, driver extensions have just been released to deliver functional improvements for owners of old ATI Radeon R300 GPUs. Linux-centric site Phoronix notes that the new OpenGL Extensions from the open-source community will also deliver benefits to X700 / X800, R400, and X1000 R500 series graphics cards.

Let’s talk about the depth of support on offer here. Old ATI X300 series GPUs first walked the Earth in 2002 AD. They were the backbone of the Radeon 9000 series graphics cards, starting with the game-changing Radeon 9700 PRO, which spearheaded the latest DirectX 9 and AGP 8X performance technologies on the desktop. For a major episode of nostalgia, flick through our 29-page review of this Red Team challenger that “proved to be superior in all possible categories.”

It’s pretty heartwarming that these veteran slivers of silicon are still carefully tended for by the Linux community. According to our searches, mainline Windows support for R300 GPUs ended with the Catalyst 9.3 unified driver in March 2009. When AMD released Catalyst 9.4 the following month, nothing older than the Radeon HD 2000 series (R600 GPU family), which debuted in mid-2007, was supported.

So, what’s new with the Linux driver? As part of the upcoming release of Mesa 25.3, the R300 Gallium3D driver will now support two memory-related OpenGL extensions that weren’t implemented prior to this week.

A new book features some of the most striking examples of graphics card box art from the late 1990s through to the early 2010s. Overclocked: An Archive of Graphics Card Box Art could be the perfect coffee table book for PC hardware aficionados, and will equally delight and/or horrify those who pick it up, depending upon their sensibilities.

A pretty good idea of the publication’s contents is provided by the cover of this tome, contrary to the old adage. It promises to be a real page-turner, and opening the book, you will enjoy the fascination-horror of “over 300 retail boxes from the late 1990s through to the early 2010s,” explains the publisher.

On the cover is a typical example of a mythical elfin enchantress, a recurring theme of the age. However, much more bizarre images can be found inside. Graphics card stalwart Gainward looks like a relatively frequent offender in the eye-bleach artwork category. Samples from this Taiwanese AIB frequently featured imagery inspired by themes like Terminators, but other brands of cyborgs were obviously available.

MSI products are also quite prevalent in this book, and it appeared to favor armor and gem-encrusted heroic figures. Sapphire is another company that remains a going concern, but has a strong history of producing shudder-inducing GPU carton artwork.

Viewers of the book may also enjoy a sense of nostalgia with artwork from brands that are now long gone, and seeing packaging for ancient GPU families.

We searched through the Tom’s Hardware archives for additional samples, but unfortunately, our old reviewers rarely took photos of the packaging artwork, or if they did, only low-resolution images were available. Nevertheless, we found enough to put together the collage above.

It was surprising to see that Amazon sites around the world seem to be one of the better sources of old GPU artwork. On the retail giant's German site, you can still see listings for the legendary ugly alien Sapphire Radeon X550 256 MB. Meanwhile, in the U.S., the product page for the Radeon HS 7770 GHz Edition is still live with its camouflage ‘busty army babe’ artwork.

Another noteworthy find from our digging was the “world's first GPU-based graphics card” from Asus, with its dreadful box art reproduced below.

AMD is reportedly prepping yet another 'GRE'-badged graphics card, but this time, it will belong to its outgoing Radeon RX 6000 series. Suppose the information from Hoang Anh Phu, a PC hardware leaker, is accurate. In that case, AMD's Radeon RX 6750 GRE carries the Navi 22 graphics processor from Radeon RX 6700, albeit with higher clocks, to compete with the best graphics cards.

AMD's Radeon RX 6750 GRE is essentially a version of the Radeon RX 6700 with 2,304 stream processors running at a high frequency (i.e., higher than 2,450 MHz recommended by AMD for the Radeon RX 6700) and featuring 10 GB of GDDR6 memory mated to the GPU using a 160-bit memory interface, the leaker asserts. For now, the actual Radeon RX 6700 GRE clocks are unknown, but if they are high enough, the product could offer an appreciable compute horsepower over the regular Radeon RX 6700.

The Radeon RX 6750 GRE likely has the same memory subsystem as the Radeon RX 6700. The Radeon RX 6700 XT has 12 GB of memory connected to the graphics processor using a 192-bit interface and features a peak bandwidth of 432 GB/s. By contrast, the Radeon RX 6700 features a peak DRAM bandwidth of 384 GB/s. It's uncertain if the Radeon RX 6750 GRE will have faster memory.

Boosting GPU performance while cutting down memory bus may seem a bit illogical, but this is not something we have not seen before. ATI Technologies, which is now known as AMD's Radeon Technologies Group, had Radeon 9500 Pro and Radeon 9500 XT back in 2002; the former featured eight-pixel shader pipelines and a 128-bit memory interface, whereas the latter boasted a 256-bit memory bus albeit only four-pixel shader pipelines.

We have no idea about the price or availability timeframe of AMD's alleged Radeon RX 6750 GRE. In fact, given that we are dealing with unofficial information, we would take it with a grain of salt in general. Furthermore, the Radeon RX 6750 GRE may be released in select countries for system integrators, so it will be hard to get it in retail.

AMD and ATI Technologies ULC have filed a complaint with the United States International Trade Commission (USITC) that Realtek Semiconductor and TCL Industries Holdings have violated five patents, triggering a USITC investigation of the claims.

The USITC announced that it would investigate certain graphics systems, components thereof, and digital televisions with components from Taiwan-based Realtek and the China/Hong Kong-based TCL Industries Holdings (and its subsidiaries). The allegedly patent-infringing products are all imported into the US for sale, bringing them under the remit of the USITC.

AMD and ATI filed the complaint on May 5, 2022, seeking an exclusion order and cease and desist orders for the sale of the products. The USITC announced the investigation on July 1. AMD and ATI Technologies ULC of Canada, which AMD purchased back in 2006, claims Mediatek and TCL Industries violated five patents covering various graphics technologies. 

The ATI patents include texture decompression techniques, a graphics processing architecture with a unified shader, and a multi-threaded graphics processing system (patents 7,742,053 claims 1-9, 8,760,454 claims 2-11, and 11,184,628 claims 7-12). The AMD patents cover a method and system for synchronizing thread wavefront data and events, and a patent covering a processing unit that enables asynchronous task dispatch (patents 8,468,547 claims 16-21, and 8,854,381 claims 15-20). 

The investigation brings back memories of another USITC investigation triggered by AMD and ATI back in 2017. That investigation centered on patent infringements by LG, Vizio, Mediatek, and Sigma Designs, most of whom settled with AMD. However, AMD eventually sued Mediatek for violating several of its patents. 

AMD and ATI's complaint claims that Realtek and TCL Holdings violated section 337 of the Tariff Act of 1930. The first step of the 'Section 337 Investigation' is for the Chief Administrative Law Judge of the USITC to designate a presiding Administrative Law Judge to oversee an evidentiary hearing to determine if there has been a violation of Section 337. The USITC will give a target date for the completion of the investigation within 45 days.  

If Realtek and TCL are found to be in violation of AMD/ATI's patents, we can expect them to end up paying licensing fees for using the tech — that is if the case follows the traditional trajectory, of course. We've reached out for comment from the companies involved and will update as necessary. 

AMD's Markham, Ontario-based office is looking for a system-on-chip verification engineer to work on a next-generation complex SoC. Developers from Canada typically work on numerous projects involving graphics processors, including GPUs and console SoCs, which may mean that AMD is already gathering engineers for next-generation consoles from Microsoft and Sony. Or is it? We have some doubts. 

"The team behind the chip powering Xbox, PlayStation, and the latest RDNA-family graphics chip is hiring for its Markham location in Canada for the next-generation chip development project," AMD's job posting over at LinkedIn reads.  

In AMD's classification, all of its APUs, CPUs, GPUs, and custom SoCs are designated as system-on-chips, which is correct from an engineering point of view since modern processors carry a lot more than just application-specific circuits. Therefore, given the historical specialization of the Markham, Ontario-based office (former headquarters of ATI Technologies, a developer of GPUs and console SoCs), we are indeed talking about the development of two types of SoCs — GPU SoCs and console SoCs.  

Meanwhile, since the job posting mentions Xbox and PlayStation SoCs ahead of GPUs, this clearly leads to speculation that AMD is already working on system-on-chips for Microsoft's Xbox Next (5th Generation Xbox) and Sony's PlayStation 6. However, this assumption may be incorrect. 

SoC verification procedures involve verification of design compliance with planned specifications (which includes performance, thermal, electrical, and other characteristics) before the design is finalized and sent out for manufacturing. SoC verification process is a relatively late phase of chip development, so if AMD is looking for a verification engineer for a console SoC now, it means that this SoC has been in development for two or three years and is due to be mass-produced in 2024.  

It usually takes about four to five years to build an all-new CPU or GPU from scratch. So we seriously doubt that Microsoft and/or Sony ordered development of its next-generation console SoCs in 2019 or even 2020 without hearing feedback about their next generation hardware (PlayStation 5 and Xbox Series X|S) from game developers. Even if AMD is looking for an engineer to work on a console SoC now, that console SoC will hardly power a next-generation console but rather a mid-cycle update machine. Perhaps, Microsoft and/or Sony want to offer improved version(s) of their XBX and PS5 system(s) in circa 2024 ~ 2025 or just radically cut down their costs, which is why they need a new SoC. 

Meanwhile, keeping in mind that in AMD's classification a GPU is also an SoC, the Markham, Ontario-based office may just be looking for a verification engineer who is going to work on a chip due in 2024 (RDNA 4, anyone?). 

"We are currently looking for a System-on-Chip Verification Engineer who will be part of a team working on next generation of a complex SOC design," the job posting continues. "The successful candidate will play a key role in SOC verification performing the following duties for functional, power, and performance aspects with simulation and hardware emulation environment."

Intel Graphics has announced that it has teamed up with Finland's Siru Innovations. The collaboration brings in "decades of experience developing graphics IP and software services," which could be instrumental in Intel maximizing its chances of successfully establishing this relatively new division. It's all part of the ongoing ramp up of Intel Arc Alchemist and beyond, slated to arrive on desktops in the near future.

Those familiar with Siru's previous projects might be just as impressed with the tie-up as Intel graphics boss Raja Koduri. Today, Koduri seemed positively bubbly as he retweeted the Intel Graphics announcement, running through a list of notable projects past from Siru's heritage.

Siru Innovation was officially established as a company in 2011. However, its roots and technical pedigree are much deeper. As Koduri highlights in his tweet, some of Siru's core talent came from a demo group called Future Crew. Cue the 80s chiptunes...

Future Crew was established in 1986 as a Commodore 64 demo group, but moved quickly to PC demos in 1988. Influential demos with advanced 3D for the time included Unreal (1992), Panic (1992) and Second Reality (1993). Members of Future Crew split and found careers in PC gaming and companies like FurtureMark, Remedy, and the hardware company Bitboys.

Bitboys was founded in the early 90s and, after some controversy about a new graphics card IP that never materialized, it was acquired in 2006 by ATI Technologies. This same team went on to work for Qualcomm in 2009.

That leads us nicely up to 2011 and the establishment of Siru Innovations as an enabler of next gen products and provider of SoC design skills and associated software development. It says it has "designed, implemented, and verified major architectural innovations to our customer's GPUs which are in shipping products," and specializes in low-power designs. Moreover, on the software side, Siru says it has a deep understanding of computer graphics APIs and architectures down to the finest details.

The Intel Graphics tweet says that Siru is going to be a great help in gaming graphics, as well as in the fields of ADAS, hyperscale computing, and more. Sadly, no specific details are provided about what Siru will be doing for Intel. Moreover, Intel Graphics doesn't provide an outline of the new relationship with Siru, other than saying "welcome aboard." Are they being hired as consultants, or acquired and brought in house? We will update if/when we hear about the details of the relationship.

Nvidia's GeForce RTX 3090 Ti promises to vie for a spot on the best graphics card for gaming when it becomes available in a week or two, but this card will require a lot of power and a lot of cooling. Early pictures of GeForce RTX 3090 Ti graphics cards from Nvidia's partners show massive cooling systems that are 3.5 – 4 slots wide.

Nvidia's GeForce RTX 3090 Ti reportedly uses a fully-fledged GA102 GPU with 10,752 CUDA cores (up from 10,496 on the GeForce RTX 3090) mated with 24GB of GDDR6X memory running at 21Gbps across a 384-bit interface. The unit has all the chances to become the world's fastest graphics card, but this also means a rather extreme power consumption. Reports indicate that GeForce RTX 3090 Ti will have a recommended thermal board power (TBP) of around 450W. Still, some of Nvidia's partners may go as high as a whopping 480W (it is better not to think how much such a board will cost and keep in mind that Nvidia's current GeForce RTX 3090 has a power rating of 350W).

But 450W – 480W or thermal power needs some excellent cooling, and with air cooling, things get complicated. Nvidia's GA102 is a highly complex GPU, and with its 28.3 billion transistors, it wants power. But GDDR6X memory with its PAM4 encoding also consumes a lot of power, so the card emits vast amounts of heat.

This situation is where rather large cooling systems come into play. VideoCardz has gathered images of Nvidia's GeForce RTX 3090 Ti partners cards (including those from Colorful and EVGA); they are either 3.5 – 4 slots wide or hybrid liquid cooling.

A quad-slot cooling system is, without any doubt, remarkable. Yet it reminds us of some other graphics subsystems we have encountered before.

The year was 2006, and the battle for the best graphics subsystem between ATI Technologies and Nvidia was raging. It was then possible to increase graphics performance by adding GPUs into a subsystem, so both Nvidia and ATI introduced their SLI and CrossFire 2-way GPU technologies. But this was not enough for Nvidia, which thought there was some space for ultra-high-end graphics subsystems involving four GPUs.

To make those 4-way SLI possible, Nvidia developed its dual-GPU GeForce 7900 GX2 (for PC OEMs, for regular users, it followed with GeForce 7950 GX2). Quad SLI did not work correctly (yet the interest towards that setup was extremely high), but it set the tone for extreme performance and power consumption.

Nvidia's GeForce 7900 GX2 consumed some 110W of power and required a hefty power supply unit (by the standards of 2006), and two of such cards needed 220W of power. Today, a single chip with its top configuration running at high clocks can use some 450W or more. Funny, this power consumption with a single graphics processor is comparable to that of four GPUs back in the day.

Then again, Nvidia's GA102 is faster and more capable than nearly every GPU designed. So maybe it is worth it?

AMD on Monday completed the acquisition of Xilinx, creating a company that can offer various types of compute devices, including CPUs, GPUs, and FPGAs. The move will help AMD to continue expanding its presence in the datacenter sector and offer unique solutions that will combine IP ingredients designed by the two companies. Interestingly, the first fruits of the deal are expected to materialize next year. 

AMD CEO Dr. Lisa Su told analyst Patrick Moorhead that the first processor combining Xilinx technologies will arrive in 2023, which is in contrast to the company's previous integration efforts. After AMD bought ATI Technologies in 2006, it took the company five years to build its first accelerated processing units (which included AMD's x86 cores and ATI's GPU). This time AMD inked a long-term development pact with Xilinx and was able to work collaboratively even before the regulators approved the transaction. It remains to be seen what exactly AMD plans to offer, but it is reasonable to expect the new processor to feature AMD's x86 cores and Xilinx's programmable engines.  

The merger between AMD and Xilinx is one of the cases when the combined company represents more than the sum of its parts. AMD sells CPUs for client PCs and servers, graphics processors for client PCs and HPC/datacenters, semi-custom system-on-chips for game consoles, and semi-custom SoCs for embedded applications in the automotive and entertainment industries. AMD's products rarely overlap with those from Xilinx, which are used in aerospace, analytics, AI inference, automotive, communications, datacenter, HPC, industrial, and many other verticals.  

There are certain industries when AMD's high-performance CPUs or GPUs compete against Xilinx's FPGAs or special-purpose ASICs, but in most cases, AMD's CPUs and Xilinx chips/software can complement each other in one system. Now that AMD can offer CPUs, GPUs, FPGAs, and adaptive SoCs, its commercial platforms can become significantly more competitive. 

 "The acquisition of Xilinx brings together a highly complementary set of products, customers and markets combined with differentiated IP and world-class talent to create the industry's high-performance and adaptive computing leader," said Lisa Su, chief executive of AMD. "Xilinx offers industry-leading FPGAs, adaptive SoCs, AI inference engines and software expertise that enable AMD to offer the strongest portfolio of high-performance and adaptive computing solutions in the industry and capture a larger share of the approximately $135 billion market opportunity we see across cloud, edge, and intelligent devices." 

Of course, it will take time to integrate Xilinx into AMD and build platform offerings comprising both companies' hardware. But even the journey of a thousand miles begins with the first step, which is what the two companies formally start today. Surprisingly, the two companies expect quite rapid progress in developing products encompassing IP from AMD and Xilinx, which is a result of collaborative work that spanned for well over a year. Meanwhile, AMD's and Xilinx's roadmaps will not be affected by the transaction, so all the products previously announced in one way or another will be developed and will come to market.

The Xilinx business will become AMD's Adaptive and Embedded Computing Group (AECG), led by former Xilinx CEO Victor Peng. As a result, Xilinx will maintain its leadership for at least a while. Furthermore, AMD's embedded business will cease to be a part of the company's enterprise and semi-custom unit and will merge into AECG, which might be good news as executives from the enterprise division will now spend more time on EPYC CPUs. 

Qualcomm on Tuesday said that its upcoming Snapdragon system-on-chips (SoCs) for PCs that are set to include Arm cores designed by the Nuvia team will set the performance and efficiency benchmark for Windows PCs and challenge processors from Apple and Intel. The new SoCs are set be sampled with PC makers in August 2022 and will be launched commercially in 2023.

Qualcomm has been offering its Snapdragon platforms for notebooks since 2017 and has done a lot of work enabling the Windows-on-Arm ecosystem together with Microsoft. But despite all its efforts, it has not gained any significant market share. That's partly because not all Windows programs work perfectly on Arm-based systems, and partly because the performance Qualcomm's platforms offer is lower compared to x86 platforms as well — not to mention Apple's Arm-powered machines. Qualcomm hopes that the $1.4 billion acquisition of Nuvia, its design team, and microarchitectures will finally make it a stronger player on the PC market.

Up until now, Qualcomm essentially offered beefed up versions of its smartphone-oriented Snapdragon SoCs to PC makers. That's a viable strategy to address a specific segment of the market, but it's not good enough to compete against products from Apple, AMD, and Intel. With its next-generation Snapdragon for PCs, the company will offer SoCs designed specifically for PCs from the ground up. Qualcomm said the custom Nuvia CPU cores will be tailored for personal computer workloads and the Adreno GPU will be scaled to the level of standalone graphics processing units. It did not elaborate on how precisely it will do this.

Building a comprehensive integrated GPU could be more challenging than building a high-performance CPU core. Qualcomm's Adreno team includes loads of engineers from ATI Technologies and AMD who have experience with building high-end discrete graphics processors. However, these GPUs tend to be large and power hungry, and SoC developers have to come up with solutions that offer a balance between performance, die size, and power.

For example, Apple's M1 Max has an integrated GPU that offers performance sometimes akin to that of Nvidia's mobile GeForce RTX 3060. That's a good result, particularly for an integrated solution, but it's not consistent. In some tests, we've seen performance that's closer to half of what you'd get from a desktop 3060. Either way, there are also users who demand something faster.

Qualcomm's message essentially says that with its upcoming Snapdragon generations, it will bifurcate its mobile and PC SoC development in a bid to deliver the best hardware possible. The company still stresses that Nuvia microarchitectures will be opportunistically extended to mobile, automotive, and data centers, so at some point Nuvia's technologies could be used for smartphones, but for now the plan is to build a high-performance client PC SoC that will defeat or at least challenge offerings from Apple, AMD, and Intel.

Qualcomm expects to deliver the first samples of its next-generation Snapdragon SoC for notebooks with Nuvia general-purpose CPU cores next August and release it commercially in 2023, the company said at its Investor Day 2021 summit. It has not yet indicated what process node the chips will use, but a 5nm class design seems likely.

After radically changing its approach to GPUs and hiring Raja Koduri in 2017, Intel has been rather aggressively hiring GPU engineers from various companies in general and from Raja Koduri's former company, AMD, in particular. Last month the company poached Vineet Goel, a graphics technology veteran from AMD.

At Intel, Vineet Goel will serve as Vice President and General Manager of Xe GPU Architecture and IP Engineering at Intel and will be responsible for development of the company's Xe architectures. An Intel spokesperson confirmed to CRN that Goel will head a group of architects and engineers that will be 'architecting, designing and verifying Intel’s Xe IP road map.' Since the first two or three Xe architecture generations have already been defined (or at least named) Goel will have an impact on Intel's Xe GPUs that are years down the road. 

Vineet Goel is has had quite a career in the graphics industry. Most recently he served as Corporate Vice President of GPU Architecture at AMD and before that (from August 2016 to June 2019) was responsible for GPU architecture development at the company. Given the timeline, Vineet Goel played a key role in development of AMD's latest RDNA and RDNA 2 as well as CDNA and CDNA 2 architectures. Furthermore, he also contributed strongly to development of upcoming RDNA 3 and CDNA 3 architectures as well as their successors.

Goel he started his career at Real3D in the late 1990s and continued it at ATI Technologies Orlando division after Real3D was sold to Intel in 1999. He worked at ATI and then at AMD till 2011, when he joined his former colleague Eric Demers at Qualcomm and worked on ultra-low-power Adreno GPUs for Snapdragon system-on-chips. He spent about five years at Qualcomm and then re-joined AMD in mid-2016.

Over his career Goel was granted dozens of graphics related patents, but avid readers might remember him for his work on ATI's TruForm and other tessellation technologies (for Microsoft's Xbox 360, ATI's Terascale/Terascale 2 architectures) supported by the company's GPUs over the years. His more recent inventions include FidelityFX Super Resolution, multi-instance GPUs, and chiplet-based machine learning accelerators. 

Vineet Goel is a yet another ex-AMD graphics engineer or executive that Intel has hired in the recent years. The company has also hired multiple people from Nvidia. Among Intel's noteworthy hires in the recent years are  Masooma Bhaiwala, Tom Petersen, Jon Carvill, Antal Tungler, Darren McPhee, Devon Nekechuk, Damien Triolet, and many others. Intel also recently boosted its game developers relations team with four highly-experienced executives.

File this one under ‘plenty of salt, please, waitress.’ An unverified Twitter account calling itself ‘3dfxofficial’ is making some very big claims about a return for much-missed GPU manufacturer 3dfx Interactive, last seen being absorbed into Nvidia as fast as the intellectual property lawyers could manage it, on August 5th.

Despite being aesthetically similar to 3dfx branding of the past, the artwork accompanying the tweet can be traced back to DeviantArt and an artist called vermaden.

As much as we want to believe this is real - and not just someone with an internet connection, a basic knowledge of social media, and a wry sense of humor determined to tickle the nostalgia glands of PC gamers of a certain age - we feel the chances are low. Developing a new GPU architecture is hard, as Intel knows, and doing so silently would have taken billions of dollars. Still, is a special edition GeForce card with 3dfx branding too much to hope for?

For those born this century, 3dfx was a San Jose-based manufacturer of graphics cards whose Voodoo brand was well-regarded around the turn of the millennium. Unfortunately, the company filed for bankruptcy in 2002, and its trademark was canceled in the US in 2007. 

The company pioneered GPU features such as hardware transform and lighting, and MPEG decoding, but its consumer products were outclassed by early Nvidia boards such as the Riva TNT. Its final board, the quad-processor Voodoo 5 6000, was thought capable of taking on Nvidia’s GeForce 3 and ATI’s Radeon 7500 but suffered from a bug so severe it halved its AGP speed down to 2x and broke compatibility with Pentium 4 motherboards.

We have contacted Nvidia for comment and will update this article if it arrives. 

AMD's Radeon RX 6900 XT Liquid Cooled is not supposed to be available to general audience, but only to PC builders. Retailers in India are offering this product without the need to purchase a pre-built system. The price dramatically exceeds that of AMD's MSRP for an air-cooled Radeon RX 6900 XT, one of the best graphics cards in its own right.

AMD officially said that its Radeon RX 6900 XT Liquid Cooled was a limited edition product set to be available only to select system integrators (SIs). The part about SIs is important because it indeed means relatively low volume PC builders as opposed to original equipment manufacturers (OEMs), which are high-volume PC suppliers. 

Apparently, three retail stores in India have Sapphire's Radeon RX 6900 XT Liquid Cooled in stock. The price is approximately $3,000 (which includes India's sales tax). While the price looks extremely high even by today's high prices for graphics cards, it is indeed a very high-end product that has a number of advantages, such as 16 GBs of 18 GT/s GDDR6 memory (a regular version carries 16 GT/s GDDR6 SGRAM). 

Sapphire, which is at least partly owned by Chinese contract maker PC Partner, is a historical manufacturing partner of ATI Technologies (now owned by AMD). The fact that AMD's reference cards are built by PC Partner is well known, yet not discussed widely. To that end, it is surprising to see a Sapphire-branded Radeon RX 6900 XT LC card.

Will AMD's Radeon RX 6900 XT Liquid Cooled be made widely available? Only time will tell.

In a partnership with Nvidia, Colorful has opened the world's first graphics card museum. The museum, which is located in Shenzhen, China, is expecting to open its doors to visitors soon.

Colorful recently relocated its headquarters to Shenzhen’s New Generation Industrial Park. The brand might not be well-known on this side of the globe, but Colorful is one of the oldest players in the Chinese market.

There are many rare graphics cards on display at Colorful's museum hailing from the '80s that you won't find on anyone's Best Graphics Cards list today. The company has each one categorized chronologically. The list of graphics cards includes old-school Voodoo graphics cards and Nvidia's GeForce 256, dubbed the world's first graphics card. Some of the chipmaker's first GeForce gaming graphics cards are also part of the exhibition. 

Not everything is about Nvidia though. Colorful also has a priceless collection of ATI graphics cards, such as the legendary Rage Fury MAXX. It was the company's first dual-chip graphics card and binded two Rage 128 Pro chips together on the same PCB with ATI's alternate frame rendering (AFR) technology serving as the main highway for communication. The museum also houses other rare artifacts from IBM, 3Dlabs, Intel, S3, Trident, 3Dfx and many others.

Since it's Colorful's museum, after all, the company also has diverse sections to pay homage to the evolution of eSports in China, as well as the brand's own iGame and Kudan bloodline.

Colorful also endowed its museum with some of the latest toys on the market. The company has setup a racing simulator with three 8K displays and a VR station.

Colorful stated that the museum will be open for visitor registration "soon," but didn't commit to a specific date.

A group of enthusiasts have developed a Mini-ITX motherboard carrying the MCST Elbrus-8CB 8-core processor developed in Russia and based on a proprietary VLIW-like architecture. The platform is designed for various applications that require ultimate security as well as enthusiasts who want to try the unique CPU. The crew is currently crowdfunding money for production of the IcepeakITX Elbrus-8CB motherboard.

The Only General-Purpose VLIW CPU Family Today

General-purpose processors based on very long instruction word (VLIW) architectures have never taken off to a large degree because they are hard to program for. Intel tried VLIW with its Itanium (IA-64) processors for mission-critical applications and ultimately had to abandon them after its x86-64 Xeon processors gained generally similar capabilities and much higher performance. ATI Technologies (and eventually AMD) used a VLIW architecture for its TeraScale family of GPUs, but moved on to RISC-like architectures with GCN and its successors starting early 2010s as it wanted to make its GPUs more suitable for general-purpose compute workloads.  

Today VLIW architectures are mostly used for DSPs as well as media processors for consumer electronics. But there is one company that still tries to polish off a VLIW architecture, it is called Moscow Center of SPARC Technologies (MCST) and it has been developing its VLIW-based Elbrus processors since 1986.  

The Elbrus processors are based on a proprietary microarchitecture, but can also emulate x86 and x86-64 via runtime binary translation if needed. In the recent years MCST developed a number of server-grade Elbrus processors with up to 16 cores primarily for machines used by the government. 

The IcepeakITX Elbrus-8CB: A Universal Motherboard

Since MCST's Elbrus processors feature a rather unique VLIW architecture, a number of desktops based on these processors aimed at software developers were introduced in Russia in the recent years. But those machines could not really be called high-end or universal even though some of them relied on high-end versions of Elbrus CPUs with eight cores.  

By contrast, the IcepeakITX Elbrus-8CB is designed for systems that could be used for a wide variety of applications that need performance, security, and ultimate reliability. Essentially, the mainboard enables PCs that are more than desktops for software makers. 

The IcepeakITX Elbrus-8CB platform carries an eight-core MCST Elbrus-8CB 1.50 GHz processor (which can be cooled down using any LGA3647 heatsink) that can be paired with 8 GB or 32 GB of DDR4-2400 memory, a PCIe 2.0 x16 graphics card, up to two M.2-2280 SSDs with a SATA interface as well as up to four 2.5-inch or 3.5-inch SATA storage devices. The motherboard comes with the MCST KPI-2 south bridge and carries multiple third-party controllers to support four GbE connectors (three RJ45, one SFP), two HDMI 2.0 display outputs, two USB 3.0 as well as five USB 3.0 ports, a COM header, audio connectors, and GPS. 

To ensure ultimate reliability, the IcepeakITX Elbrus-8CB uses a 14-layer PCB. In addition, the platform has a TPM SPI connector, a temperature sensor, two tampering sensors, gyroscope, fall detection sensor, and a water sensor. The processor and chipset use custom copper heat spreaders to ensure better heat dissipation. 

Coming Soon

Development of the IcepeakITX Elbrus-8CB motherboard is about to be finished, which is when the creators will contact interested parties that can sign up for updates without paying anything. Those who eventually buy the platform will also get the full source code for board schematics, design specifications, and internal design comments. Also, they will get MCST’s proprietary C/C++ compiler and bootloader, but not their source code as designers of the board do not have access to these programs. 

The latter fact actually indicates that the motherboard is not developed by MCST itself or any party close to it. To that end, it remains to be seen whether a third-party can procure processors and chipsets that are generally intended to be used by various government agencies and which are not in mass production just yet.

What to Expect?

One of the questions is what to expect from the Elbrus-8CB processor and the aforementioned IcepeakITX Elbrus-8CB platform. The CPU features eight cores operating at 1.50 GHz, 512 KB L2 cache per core, 16 MB unified L3, a quad-channel DDR4 memory controller (it is unclear whether the board uses all four channels) that supports up to 128 GB of DRAM, and a proprietary I/O channel offering 16 GB/s of bandwidth. The chip packs 3.5 billion transistors and is designed to be made using TSMC's 28 nm fabrication process. The chip dissipates up to 90 W of power. 

The CPU uses the 5th generation Elbrus architecture and can deliver up to 24 FP64 FLOPS per clock per core when executing a program compiled for Elbrus, which is higher when compared to AMD's EPYC 'Rome' processor that can retire 16 FP64 FLOPS per clock per core. Yet, at 1.50 GHz, the CPU is capable of ~576 FP32 GFLOPS and ~288 FP64 GFLOPs, which is significantly lower when compared to AMD's 64-core EPYC CPUs (up to 2.60 GHz ~ 3.40 GHz) that can hit around 2.3 FP64 TFLOPS, nearly ten times higher. Meanwhile, modern eight-core AMD EPYC CPUs are clocked at up 3.70 GHz ~ 3.90 GHz and will therefore beat MCST's offering. 

MCST says that its runtime binary translation for x86/x86-64 works so efficiently that actual performance impact is not significant. Yet, it is apparent that the eight-core Elbrus-8CB CPU clocked at 1.50 GHz running x86-64 emulation will be considerably slower than its eight-core x86-64 counterparts executing the code natively and running at higher clocks. 

Of course, the Ebrus-8CB has an indisputable trump over its x86 counterparts. There are no (or virtually no) viruses developed for the proprietary VLIW architecture. In addition, the CPU has a number of integrated security capabilities. Therefore, if someone wants to build a computer with advanced built-in security, the IcepeakITX Elbrus-8CB platform is an option.

The Price

There is no word on expected pricing of the IcepeakITX Elbrus-8CB motherboard, but since we are talking about proprietary hardware that will not be produced in huge quantities, the device will be priced accordingly. Earlier this year MCST itself unveiled two Mini-ITX motherboards based on its Elbrus-1C+ and Elbrus-8C1 processors (i.e., previous-generation chips) priced at 92,000 rubles ($1,204) and 120,000 rubles ($1,572), respectively, though it is unclear whether the price includes VAT.

Sapphire Technology has posted a teaser of its upcoming custom Nitro+ Radeon RX 6800 XT graphics card. The board is equipped with a rather remarkable cooling system that promises to enable higher-than-AMD-recommended frequencies out-of-box as well as some additional overclocking potential. 

Since AMD designed a rather compelling triple-fan cooling system for its Big Navi-based family, many graphics card vendors — Sapphire included — decided to go with reference coolers and cards with their initial Radeon RX 6800/6900 offerings. But custom coolers with better performance enable makers to build factory-overclocked graphics cards that are sold at a premium, so Sapphire and its rivals are eager to bring them to market. 

The Sapphire Nitro+ Radeon RX 6800 XT graphics card comes with a rather interesting 2.5-wide or even 3-wide cooler that that has three fans of different sizes. One of the fans looks particularly larger than two others. The powerful custom cooling system almost certainly means a custom PCB with an enhanced VRM too, so expect Sapphire's Nitro+ RX 6800 XT to offer increased stock clocks along with some additional overclocking potential. 

Unfortunately, Sapphire did not disclose when it plans to release its Nitro+ Radeon RX 6800 XT as well as its out-of-box frequencies. Though, it is reasonable to expect this card to arrive in the coming weeks. 

The Nitro+ Radeon RX 6800 XT certainly looks impressive and could easily feature one of the most advanced air-cooling solutions around. In the meantime, it remains to be seen what Sapphire has for its top-of-the-range Radeon RX 6900 XT offering up its sleeve.  

Back in the day Sapphire was one of ATI's (and then AMD's) key graphics cards partners and had a very special treatment, so the company experimented with off-the-shelf cooling solutions quite often. It was first with a 'liquid metal'-based cooler in 2005, then it was first to offer a stock liquid cooling system with a graphics card in 2006, and then it was first to offer a vapor chamber cooler in 2008. I

AMD has reportedly sued MediaTek for allegedly violating several patents. MediaTek stands accused of using technologies covered by those patents, which are related to AMD's accelerated processor units (APUs) and GPUs, in products ranging from television sets to unidentified "smart devices."

The report was originally published by Bloomberg on its wire service; a summary is available from Seeking Alpha. AMD is said to be seeking "cash compensation for past, continuing, and future infringement" as well as a court order to block future infringement of its APU and GPU patents. The company is sending a clear message--don't use its graphics tech in your products if you aren't willing to license the associated patents first.

This lawsuit follows a complaint AMD filed with the U.S. International Trade Commission (ITC) in 2017 against MediaTek, Sigma Designs, and other companies. AMD's complaints are the same as they are now: companies violated several patents related to its graphics technologies in their quests to offer cheap-but-good TVs. The ITC decided in AMD's favor in August 2018, giving the company the legal backing it needed to file a suit like this one.

Here's how we described the patents in our initial coverage of the ITC's decision:

"Three patents were at the center of AMD's complaint. Two resulted from its acquisition of ATI Technologies in 2006; the other was AMD's from the start. The patents broadly covered a parallel pipeline graphics system and unified shader technologies as well as the architecture required to use them. AMD wanted the technologies for its graphics cards, but they could also be used to power displays for other types of products, too."

It's not clear why AMD waited to file suit against MediaTek after the ITC's decision. Perhaps the companies were in settlement talks--AMD settled with LG before the ITC even issued a decision--that fell through. Either way, it wouldn't be surprising to hear sometime in the next few months that MediaTek decided to settle instead of battle AMD in court. Legal battles are expensive, and especially after the ITC's decision, this one will probably be hard to win.

Tahiti, Pitcairn, And Bonaire Show Up For An Encore

Back in 2000, a gentleman by the name of Brian Hentschel called my dorm room at UCLA to ask my opinion of brand names. Brian was a marketing manager at ATI, and the company was looking for something catchy to succeed its Rage family. I had owned every single Rage-based desktop graphics product up until that point, and was pumped to provide feedback on the company's next-gen nomenclature.

Thirteen years later, I cannot remember the other options ATI was throwing around, but I distinctly recall liking Radeon least of all. Clearly I have no future in marketing, because I’ve been reviewing Radeon-branded cards ever since.

With its latest generation, AMD maintains the Radeon legacy, but changes everything that comes after. According to the company’s PR team, the new naming scheme makes positioning easier—and I’d have to agree. Our own writers were mistyping combinations of Radeon HD 7990, 7970, 7790, and so on. Now, we have the high-end Radeon R9 and mainstream R7 families, which are sub-divided into three-digit models suggesting performance levels.

Say Goodbye To The Old Names And Hello To The Old GPUs

At its press day in Hawaii, two weeks ago, AMD publically announced the Radeon R7 250, R7 260X, R9 270X, R9 280X, R9 290, and R9 290X. There’s also an R7 240 the company didn’t mention. How on earth will you ever memorize all of the corresponding specifications of each card in a timely manner? It’s easy: although we’re looking at new model names, all of the products AMD is talking about today employ GPUs already found in the Radeon HD 7000-series line-up.

Take that R9 270X, for example. With 1280 shaders spread across 20 compute units, it employs the same Pitcairn GPU introduced on the Radeon HD 7870 GHz Edition in March of last year. Or how about the R9 280X? Its 2048 shaders, 1 GHz engine frequency, and 384-bit memory bus should remind you of the Radeon HD 7970 GHz Edition, sporting the Tahiti GPU.

Of course, taking existing technology, tweaking it a bit, and giving it a shiny new-sounding name is an old practice. Much of the Radeon HD 8000 family is a replicate of the 7000s, shipped off to OEMs in the hope that folks buying tier-one machines don’t know any better. And don’t think I’m picking on AMD here. Nvidia’s GeForce GTX 700M series has Fermi-based models in it with core configurations dating back almost three years. The GeForce GTX 770 and 760 employ the same GK104 GPU found at the top of the 600-series. This sort of thing seems to happen a lot in the graphics market.

The good news for today is that familiar GPUs make our job quite a bit easier. Doubly-so because the two products based on never-before-seen silicon, R9 290 and 290X, still aren’t ready for their public debut. This leaves us with the remainder of AMD’s R9 and R7 line-ups, well-known (and tested) technology, and price drops across the board. Positioning becomes the main focus of today's discussion, then.

Just don't be quick to marginalize what AMD is doing. Most Radeon HD 7970 GHz Edition cards currently sell for somewhere around $375. R9 280X is going to debut at $300. The GeForce GTX 770 that Nvidia launched to replace its GTX 680 still sells for $400. Remember when the original 7970 sold for $550? Boy, that escalated quickly.

Let’s take a closer look at the R9 280X—for now, the highest-end board in AMD’s re-branded portfolio.

R9 280X: The Tahiti GPU’s Second (Or Third?) Lease On Life

I guess it depends whether you consider the Radeon HD 7970 GHz Edition a distinct product introduction. If so, then R9 280X is the third single-GPU graphics card sporting AMD’s complete Tahiti GPU.

When I wrote AMD Radeon HD 7970 GHz Edition Review: Give Me Back That Crown!, that card featured a base clock rate of 1 GHz that’d jump to 1.05 GHz under the effects of PowerTune with Boost. Otherwise, it employed Tahiti in its uncut form—with 2048 Stream processors, 128 texture units, 32 ROPs, and a 384-bit aggregate memory bus populated with 3 GB of GDDR5 at 6 GT/s. AMD slapped a 250 W maximum board power on the card and started shipping it at $500 (already down $50 from the original 7970’s launch).

The R9 280X’s specs are very, very similar. Its Tahiti GPU boasts the same 2048:128:32 configuration, with a 384-bit memory bus rocking 3 GB of GDDR5 that AMD says should operate at 6 GT/s. You’ll need to use the same six- and eight-pin connectors for an identical 250 W board power. The one notable difference is the 280X’s engine clock, which tops out at 1 GHz. As a result, you’ll notice the R9 280X performing slower in our benchmarks. Fortunately, there’s that price drop…

At $300, an R9 280X does battle against the GeForce GTX 760, mostly. I say mostly because the cheapest 770s are selling for around $400 (clearly too high to be in the same league), while the 760s are between $250 and $320. And since we already know the Radeon HD 7970 GHz Edition and GeForce GTX 770 are the cards that trade blows, well, AMD should be in a pretty strong position by the time we get to the end of our benchmarks.

To answer whether existing Radeon HD 7970 cards can be paired with the new R9 280X, yes, they work together. A quick Fraps-based test showed one 280X hitting 52.9 FPS in Battlefield 3 at 2560x1440. Dropping a 7970 GHz Edition next to the newer board pushed frame rates to 102.3 FPS. When it wasn't in use, the 7970 properly spun down according to AMD's ZeroCore technology.

We received two different takes on the R9 280X, neither of them conforming to AMD’s reference design.

Asus R9280X-DC2T-3GD5

The first card was Asus’ R9 280X DirectCU II TOP, an overclocked board with a GPU capable of up to 1070 MHz and GDDR5 memory at 6400 MT/s. According to Asus, it’ll sell this variant for $310, bundled with a power adapter and CrossFire cable.

Beyond the tweaked specifications, Asus employs a non-reference PCB and oversized cooling to help manage power, thermals, and acoustics. The company says AMD’s reference 280X employs a five-phase (60 A/phase) PWM, while its own card features eight phases at 45 A/phase. This, along with claimed higher-quality power components, is supposed to benefit aggressive overclocking. Naturally, Asus bundles its GPU Tweak software for adjusting core clock, voltage, memory, power target, and fan speed (mostly settings you can tune in AMD’s OverDrive applet, except for voltage).

Although the R9 280X DirectCU II TOP is a dual-slot card, it’s also long and tall. The fan shroud stretches over the back of the PCB, imparting an 11.2” overall length. Further, a heat pipe coming out of the board’s top increases maximum height to 5.7”.

Display outputs include one dual-link DVI-I port, one dual-link DVI-D connector, HDMI, and a full-sized DisplayPort ouput.

XFX R9-280X-TDFD

XFX’s R9-280X-TDFD showed up next, based on a slightly different PCB. This is the one we benchmarked for our performance evaluation, if only because we were able to get its speeds and feeds first. An 850 MHz core clock rate accelerates up to 1 GHz when the thermal headroom allows. The memory tops out at AMD’s reference 6 GT/s, serving up to 288 GB/s of bandwidth.

Otherwise, the R9-280X-TDFD is a dual-slot board employing two axial-flow fans that do their job very quietly, but dump waste heat back into your case instead of exhausting it out the back.

Display outputs total five. You get one dual-link DVI-I port, one single-link DVI-D connector, HDMI, and two mini-DisplayPort outputs. Quad-card CrossFire configurations are supported across two bridges, though an extra-tall plastic frame surrounding the heat sink makes even a flexible ribbon connector difficult to attach.

R9 270X: Pitcairn Gets A Little Boost

When the Radeon HD 7870 launched, it sold for $350. Now, you can find the cards going for somewhere between $185 and $200. Incidentally, AMD wants to introduce its R9 270X, based on an ASIC it’s calling Curacao (but is every bit similar to Pitcairn on paper), at the upper end of that same range: $200.

Although this isn’t nearly the savings story we heard from the R9 280X, AMD probably isn’t feeling pressured by the 270X’s primary competition. Nvidia’s GeForce GTX 660 with 2 GB is currently selling for about $200 as well, and it doesn’t keep up with Pitcairn (just prior to launch, Nvidia announced the 660 is dropping to $180).

Fortunately, for roughly the same price, AMD does bolster the R9 270X’s performance a little.

The Radeon HD 7870 had a Pitcairn GPU with 1280 shaders, 80 texture units, and 32 ROPs on it. A 1000 MHz core and 1200 MHz memory clock were ample for a solid gaming experience at 1920x1080, and an aggregate 256-bit bus with 2 GB of GDDR5 memory helped facilitate high detail settings.

Likewise, R9 270X brings to bear 1280 shaders, 80 texture units, and 32 ROPs. Its Curacao GPU notches up 50 MHz to 1.05 GHz, and the 2 GB of GDDR5 on our press sample runs at 1400 MHz (or 5.6 GT/s). There will be 4 GB models, AMD says, but they’ll be a bit pricier than the 2 GB version’s $200. Whereas the 7870 bore a 175 W board power, R9 270X is rated for 180 W. Fortunately, the small bump is meager enough that you’ll still find yourself using two six-pin auxiliary connectors.

Third-party implementations will likely differ in the display outputs that get exposed, but AMD’s reference model features two DVI outputs, HDMI, and DisplayPort.

R7 260X: TrueAudio’s First Outing On The Back Of Bonaire

As you saw during our coverage of AMD’s GPU14 Tech Day in Hawaii, there are three SKUs in AMD’s line-up with TrueAudio support: R9 290X, R9 290 (neither of which are available yet), and R7 260X.

The Tensilica (a fabless semiconductor company) DSPs that enable TrueAudio are built onto AMD’s GPUs, which might make you think, “Ah ha, a new feature on-die—surely we must also be dealing with a new GPU, too.” Disappointingly, no. This functionality was actually part of the Bonaire processor that launched alongside Radeon HD 7790 (AMD Radeon HD 7790 Review: Graphics Core Next At $150, from March of this year), but was simply not enabled. Now, at least, it’s available for middleware developers to play with and, eventually, expose.

Bonaire, as it appeared on the Radeon HD 7790, boasted 896 shaders, 56 texture units, and 16 ROPs. The sample we reviewed had a GPU running at 1 GHz with 1 GB of GDDR5 memory at 6 GT/s on a 128-bit bus. Power, interestingly enough, was rated at 85 W, deliverable through the PCI Express slot and one six-pin lead. Bonaire was also AMD’s first GPU with an updated version of PowerTune featuring a second-generation voltage regulation controller able to control voltage in 6.25 mV steps. The result of this functionality is much faster response to changes in input (temperature, activity, or telemetry) and corresponding voltage/clock rate/fan speed adjustments.

The GPU that shows up on R7 260X features all of the same vital specs, including the newer PowerTune implementation. Only, its engine clock operates at up to 1.1 GHz. Its memory (now 2 GB instead of 1) streaks along at 1625 MHz over an aggregate 128-bit bus, serving up to 104 GB/s of bandwidth. It’s still serviceable by a single six-pin power cable, but the result of higher clock rates all around are a board power that jumps to 115 W. That’ll hardly be an issue for most folks.

As of this writing, Gigabyte has a Radeon HD 7790 running at 1075 MHz with 2 GB of memory at 1500 MHz selling for $140 on Newegg. AMD plans to sell the R7 260X for the same $140. Most of the other 7790s are 1 GB models, so AMD is still adding some value. But there’s really not much to get excited about in the shift from Radeon HD 7790 to R7 260X, aside from TrueAudio getting switched on. And even then, we need to wait for the feature to get enabled in upcoming games.

TrueAudio: Dedicated Resources For Sound Processing

If you followed along with AMD’s tech day webcast, then you sat through a lot of TrueAudio discussion. In fact, given the amount of time dedicated to TrueAudio, the feature seemed like it’d be the day’s emphasis.

At the event, we were hearing the partner demos across eight channels, and the positional audio was certainly discernable, if not overwhelmingly busy (on purpose, no doubt). But we all know that 7.1- and even 5.1-channel sound setups outside of a home theater are very uncommon. Two- and 2.1-channel configurations, including headsets, are far more common. Unfortunately, it didn’t sound like anyone tuned in over Livestream was hearing the same output over stereo.

For anyone who was around in the late ‘90s to hear Aureal’s and Sensaura’s technologies, before both were acquired by Creative, you know that the head-related transfer functions used to create effective positional audio over two channels are not new. The point of TrueAudio is to facilitate more complex sound effects (those HRTFs aren’t computationally free) without burdening the host processor. Today, AMD says that audio gets as much as 10% of a game’s CPU utilization budget, limiting what developers can do. But with TrueAudio, AMD wants to guarantee the availability of real-time processing resources specifically for sound, and regardless of the host CPU you have installed.

This is achieved through the Tensilica HiFi2 EP Audio DSP cores mentioned on the previous page. In the R7 260X, there are two three cores integrated on the Bonaire GPU. The higher-end R9 290 and 290X will also feature three DSP cores dedicated to TrueAudio. Those DSPs employ Tensilica’s Xtensa ISA with fixed- and floating-point number support, which AMD says is equally useful for high-end gaming and embedded applications. Because the DSP is programmable by nature, you can really feed anything you want into it, so long as there’s a decoder available. To that end, the professional audio software vendors are purportedly showing an interest, eager to see what dedicated hardware can do that host-based processing couldn’t.

The real-time nature of audio in a gaming environment means that fast access to compute cycles and memory is imperative, even if the cores themselves aren't particularly powerful. Each one includes 32 KB of instruction and data cache, along with 8 KB of scratch RAM. A fast routing interface connects the DSPs to 384 KB of shared internal memory organized in 8 KB banks. The local resources are fed by a multi-channel DMA engine able to keep the cores busy. And up to 64 MB of frame buffer memory is addressable through a low-latency bus interface shared with the display pipeline.

One of the first questions that came to mind upon hearing about TrueAudio was, “will game developers, already strapped for time and money as they get their titles to market, put resources into sound when there’s so much going on in graphics, physics, and AI?” AMD seems to think that the impact on ISVs will be minimal, though. Because a majority of developers are utilizing middleware for their audio, TrueAudio needs support from those companies first and foremost. Once you get support in Audiokinetic and Firelight’s FMOD, detecting and utilizing TrueAudio becomes much easier. From there, the feature exerts its influence before getting handed off to a codec, and is consequently compatible with any output type.

What about the fact that AMD is only making TrueAudio available across three products, two of which aren’t even available yet? Representatives say that AMD has to start somewhere with TrueAudio, and this is simply the first public airing. I’d add that high-end graphics cards, destined for high-end PCs also don’t need audio effects acceleration as much as less powerful platforms. But you can guess where this is going: expect the same technology to start showing up in AMD’s APUs and mobile GPUs, which are less powerful and might even realize power benefits from accelerating audio.

Display Technology

Historically, display technology is one of AMD’s strengths. Back in 2009, the company caught its competition flat-footed with Eyefinity, which supported three independent outputs from the Radeon HD 5870 and as many as six from a special Eyefinity 6 Edition of the card. This was a pivotal moment in my career as a writer who loved to play games. Previously, I used a Quadro NVS card in my personal workstation to drive a three-screen array, while a second system handled 3D. The Radeon HD 5870 let me combine those equally important functions in one machine.

More recently, AMD was on the receiving end of an offensive because its drivers do not yield a favorable experience at 3840x2160 (frame pacing isn't supported yet in Eyefinity). But although you probably wouldn’t want to game on one or more of the company’s graphics cards at 4K resolutions, configuring a tiled display was originally easier on AMD’s hardware than Nvidia’s. Of course, Nvidia has since incorporated DisplayID 1.3 support, which automatically creates a Surround array, making the setup process that much smoother. AMD's latest software likewise streamlines usage with existing tiled panels.

The point is that AMD does support 4K TVs (30 and 24 Hz) over HDMI and DisplayPoint, and tiled monitors (60 Hz) using DisplayPort with its existing GPUs. Tiled monitors are not supported through two HDMI ports, which is how we tested Nvidia’s cards in Gaming At 3840x2160: Is Your PC Ready For A 4K Display? Frankly, that’s fine. I'd much rather connect one DisplayPort cable in MST mode anyway. We only benchmarked through the HDMI interface to facilitate video capture for our FCAT analysis tools.

Additionally, the R9 and R7 boards are making it possible to connect matching monitors to any three outputs. Previously, the requirement was that one needed to hook up via DisplayPort. If you can get your hands on a MST hub, you can even enable five- and six-screen configurations using AMD’s reference cards. The issue remains availability; the only solution comes from Club3D, and you can’t buy it anywhere in the U.S.

In the future, we have confirmation that forthcoming AMD cards will definitely support single-stream, non-tiled 4K displays as they become available and get validated. This will likely be early in 2014. It remains to be seen whether the R9- and R7-series cards getting tested today can claim the same thing. The display controller’s frequency, memory arbitration, and latency all play a role in driving that many pixels.

There’s actually quite a bit more to cover on the display technology side. But because AMD hasn’t pulled the veil off of its R9 290 and 290X cards yet, we have to hold off on that discussion. More soon, though.

Test Setup And Software

Test Hardware And Software

Results: Arma III

For each game we’re testing, we need to evaluate three different products. First up is AMD’s “new” R9 280X. As expected, it’s slower than the Radeon HD 7970 GHz Edition, though just slightly. Nvidia’s closest-priced alternative, GeForce GTX 760, sells for $50 less, but even gets beaten by the R9 270X in Arma III. The cheapest 7970 GHz Edition card (as of this writing) sells for $330, so for $30 less, the R9 280X is a good example of AMD’s Tahiti GPU made more attractive.

Stepping down one product category means giving up playable performance at 2560x1440 (at least using Very High quality settings). Nevertheless, AMD’s R9 270X has little trouble outpacing GeForce GTX 760 and the Radeon HD 7870. AMD scores a value win, without question. But with 7870s going for as little as $170, spending $30 more on an R9 270X is a step in the wrong direction, price-wise, for the same Pitcairn/Curacao GPU.

Arma is a great-looking title, and its Very High detail setting is pretty taxing. An average frame rate in the 30s might not be satisfactory at 1920x1080, compelling you to scale back on eye candy (a shame, really). R7 260X won’t change your experience compared to the Radeon HD 7790. The thing is, most 7790s are 1 GB cards. The 2 GB Gigabyte model we bought sells for the same $140 AMD plans to charge for its R7 260X. So, for the same price, you’re getting a slight overclock and TrueAudio turned on. The good news for AMD is that, even after a price drop on Nvidia’s GeForce GTX 650 Ti Boost, its Bonaire-based boards are still a better value. Our Best Graphics Cards For The Money column concurs. 

An analysis of frame rate over time at 1920x1080 and 2560x1440 breaks our 10 comparison boards into three distinct groups. Up top, the Tahiti-based offerings appear uncontested by the GeForce GTX 760, which instead competes against $200 Pitcairn-based cards.

Arma is taxing enough that, at 2560x1440, you’re probably going to want a Tahiti-class card. Otherwise, you’re going to spend a fair amount of time under 30 FPS.

In single-GPU configurations, all of these solutions demonstrate low frame time variance. For more on what this measurement includes and how we generate it, check out this page.

Results: Battlefield 3

Again, a $30 savings seems to be worthwhile, given the performance sacrifice you make going from Radeon HD 7970 GHz Edition to R9 280X. Then again, with vanilla 7970s already selling for $300 online, you didn’t exactly need to wait for a re-brand for access to Tahiti at a great price.

Although AMD fails to impress relative to its prior-generation products, it fares better against GeForce GTX 760. The R9 costs $50 extra, but is probably what you’d want in order to play Battlefield 3 smoothly at 2560x1440 using the Ultra detail preset. Really, the GeForce card is more of a match-up to AMD’s older Radeon HD 7950. But that model is going to disappear, and Nvidia doesn’t have anything else short of $400 to go up against the R9 280X at $300.

I’m not sure it’s worth spending an extra $20 for an R9 270X when the Radeon HD 7870 comes as close as it does. But you’ll probably want to choose the 7870 over Nvidia’s GeForce GTX 660, even at the adjusted $180 price point.

R7 260X’s advantage over the Radeon HD 7790 is marginal. Its lead over the GeForce GTX 650 Ti is similarly small. But the GeForce GTX 650 Ti Boost is quite a bit faster and only $10 pricier. That’s a smarter play for Battlefield 3 using the Ultra detail preset at 1920x1080.

Charting frame rate over time is good for monitoring dips into unplayable territory. At 2560x1440, you can see the R9 270X does keep minimum performance a bit smoother than the Radeon HD 7870 in a handful of passages. That could help make a case for spending the extra $30 on AMD’s higher-clocked board (or for simply overclocking your 7870).

There’s little of interest to report from our frame time variance charts. In single-card configurations, each of these solutions behaves itself.

Results: BioShock Infinite

It would have been easy to recommend AMD’s Radeon HD 7870 over the GeForce GTX 660 when it sold for $200. But a recent drop to $180 balances that price range, making it hard to declare a winner. What we do know is that, for another $20, the R9 270X doesn’t really add anything compelling to the story in BioShock Infinite.

In contrast, it’d be hard to not spend an extra $10 on a GeForce GTX 650 Ti Boost, given its advantage over the R7 260X, GeForce GTX 650 Ti, and Radeon HD 7790.

At the top end, Radeon HD 7970 GHz Edition and R9 280X serve up the best experiences at 2560x1440. To get any more, you’d have to jump all the way up to $400 for a GeForce GTX 770. This big hole in Nvidia’s line-up makes AMD’s new R9 280X the entry point for gamers looking to play demanding first-person titles at 2560x1440 using high detail levels.

The two Tahiti-based cards keep their noses above 50 FPS at 1920x1080, while Nvidia’s GeForce GTX 760 drops to the mid-40s. That’s still fast enough, if you’re playing on a FHD display. But the cheaper Pitcairn-based cards and GeForce GTX 650 Ti Boost are quick enough to keep up at 1920x1080.

Gaming at 2560x1440 is a more taxing test of each GPU’s potential. AMD’s Radeon HD 7970 GHz Edition and R9 280X maintain at least 35 FPS. Meanwhile, Nvidia’s GeForce GTX 760 spends a lot more of its time under 40 FPS, dipping closer to 30 on one occasion.

Frame time variance is a little higher in BioShock Infinite, but even at 2560x1440, the worst-case numbers don’t look too bad.

Results: Crysis 3

Nvidia's lack of a solution between $250 and $400 gives AMD an opportunity to enable playable frame rates in another demanding title at 2560x1440 with its R9 280X. GeForce GTX 760 dips close to 30 FPS, making it a better solution for cranking the details up even higher at 1920x1080.

You could also get away with a GeForce GTX 660 for $180 at that resolution instead of a R9 270X for $200. However, the smartest play is a Radeon HD 7870 for $180, as long as they're around.

At the lower end of the spectrum, Crysis 3 with a High System Setting preset is a little too demanding for R7 260X, Radeon HD 7790, or GeForce GTX 650 Ti. If this title is important to you, we again find ourselves drawn to the GeForce GTX 650 Ti Boost with 2 GB for an extra $10.

Unlike a lot of our more automated tests, Crysis 3 involves a manual run-through with live action that varies each time. As such, there’s a little more variance between benchmarks. This is most obvious from those dips you see at the very end of our 1920x1080 chart. There’s one area in the map we test that hammers frame rates. It clearly affects the R9 280X’s performance, but not the Radeon HD 7970 GHz Edition. Also not affected is the Radeon HD 7870, while the R9 270X slides down to about 30 FPS before popping right back up.

Throughout the run at 2560x1440, AMD’s fastest Tahiti-based card manages an almost-10 FPS lead over GeForce GTX 760. The Nvidia board is still plenty playable, but I’ll maintain that R9 280X is a good entry point for this resolution using the settings we’ve chosen.

Getting hammered by big performance dips at the end of our run really affects the worst-case frame time variance numbers reported by GeForce GTX 760 and R9 280X at 1920x1080. In reality, those shouldn’t worry you—the slow-down isn’t perceptible as we play through the test.

Results: Grid 2

Bigger numbers in Grid 2 mean that even mid-range cards serve up playable performance—so long as you match them up to high-end platforms with plenty of memory bandwidth. In this case, an overclocked Core i7-4960X and four channels of DDR3-1866 memory are what carry the Radeon HD 7870 and R9 270X to almost 50 FPS average rates at 2560x1440.

Tahiti justifies its price premium over the GeForce GTX 760’s GK104 at 2560x1440. The highest-end Nvidia card we’re testing, which sells for $250, barely slides in ahead of the Pitcairn-based boards. Again, Radeon HD 7870 for $180 looks like a pretty sweet deal for as long as it’s around, right?

At the bottom end, R7 260X comes in just ahead of the Radeon HD 7790, which matches its price. The GeForce GTX 650 Ti Boost, selling for $10 extra, does nothing extra for performance at 2560x1440. And its advantage at 1920x1080 isn’t significant enough to change the gaming experience.

Although performance through our Grid 2 benchmark run jumps up and down, creating fairly busy lines, we still see three clumps of cards. Unfortunately for Nvidia, its GeForce GTX 760 is part of the second clump where AMD’s cheaper Pitcairn-based cards show up.

Frame time variance is very low in Grid 2, even when we look at the worst-case 95^th percentile numbers.

Results: The Elder Scrolls V: Skyrim

Skyrim tends to be more platform-bound than most of our other benchmarks, so an overclocked Ivy Bridge-E-based configuration with lots of fast memory lets these cards perform to their peak potential using the Ultra detail preset.

The thing is, this game just doesn’t tax graphics hardware very much. You’ll still find it playable at 2560x1440, even on a Bonaire-powered Radeon HD 7790 or R7 260X. Most notable, perhaps, is that a $200 R9 270X trades blows with a $250 GeForce GTX 760.

Smooth frame rate over time line graphics demonstrate an entire field of playable performance at 1920x1080, and mostly ample numbers at 2560x1440 using the game’s Ultra quality preset.

The GeForce cards experience higher frame time variance, on average. At 1920x1080, only the 650 Ti’s worst-case result is something you’d likely notice. At 2560x1440, however, the numbers using Nvidia’s latest beta drivers aren’t as good. Again, it’s the GeForce GTX 650 Ti that demonstrates the least-favorable behavior.

Results: Tomb Raider

Radeon HD 7970 GHz Edition and R9 280X are in a league of their own through our manually-run Tomb Raider benchmark, never dropping below 50 FPS at 2560x1440. Meanwhile, the R9 270X slides in ahead of Nvidia’s GeForce GTX 760. The real winner is AMD’s Radeon HD 7870, though, which similarly does battle in the same space, but costs $20 less than the new R9.

AMD’s R7 260X is a little bit quicker than the Radeon HD 7790 it replaces, but not so much so that it’d warrant a higher price. The same holds true for Nvidia’s GeForce GTX 650 Ti Boost. Unfortunately, that card does cost more. The trio is sufficient for 1920x1080, but higher resolutions are too demanding.

To reiterate a recurring theme, the Radeon HD 7970 and R9 280X are great entry points for gaming at 2560x1440. The R9 270X, GeForce GTX 760, and Radeon HD 7870 certainly suffice as well, but they come a lot closer to marginal. Moreover, to get something faster than the $300 R9 280X, you’d have to spring for a $400 GeForce GTX 770.

Tiny frame time variances indicate a smooth experience across the board. Only Nvidia’s GeForce GTX 650 Ti runs into a bit of trouble at 2560x1440—a resolution too taxing for the frame rates to be playable anyway.

CAD: AutoCAD 2013 And Inventor 2013

AutoCAD 2013

2D Performance

The differences between cards are marginal. At this point, as long as you’re dealing with 2D output, the type of card you’re using doesn’t really matter. From workstation- to gaming-class hardware, all of these cards are sufficient.

3D Performance

There are clearer reasons to favor one board over another if you isolate 3D performance. Desktop-oriented cards do benefit from DirectX support. However, the new Radeon cards aren’t stunners by any stretch; they roughly achieve similar results as their genetic predecessors.

Autodesk Inventor 2013

Inventor also employs DirectX, which means that even gaming boards stand a chance. Interestingly, the Radeon cards fare better than they did in AutoCAD.

OpenGL: Maya 2013 And LightWave

Maya 2013

Two scenes that don’t employ the new Viewport 2.0 with DirectX support illustrate the disadvantages of consumer cards compared to workstation hardware cards as soon as OpenGL comes into play. The FirePro and Quadro drivers are simply better-optimized. But especially in the second scene, you also see that a GeForce GTX 580 does well, so long as the workload is right for it.

Lightwave

LightWave wrecks consumer cards as well. You can certainly use the R9 and R7 cards to mess with mid-sized models that aren’t too complex, but a workstation card is clearly a better option.

OpenCL: Bitmining, OpenCL, And RatGPU

Bitmining

Based on All About Bitcoin Mining: Road To Riches Or Fool's Gold?, we know that specialized hardware is much better for mining Bitcoins than graphics cards. But the practice is still worth benchmarking.

AMD leads the way, as it has in the past, though the differences between this generation and last are very small.

LuxMark

For the past two years, we’ve watched AMD dominate compute-oriented workloads. It does particularly well in the OpenCL-accelerated LuxMark benchmark, based on the LuxRender rendering system. Nvidia’s Kepler architecture isn't as inspiring for this type of task.

RatGPU

ratGPU is unique ray-tracing renderer accelerated by OpenCL and available for 3ds Max and Maya. Remarkably, the older Radeon HD 6970 dominates the rest of the line-up, while the new Radeons must be content with a mid-pack finish.

Power Consumption

Unfortunately, AMD didn’t send reference R9 280X cards to either our U.S. or German offices. Instead, we have a number of board partner designs running at different clock rates and with varying cooling solutions. We’re getting as close as possible to the reference specs, but be advised that the power consumption values might not match the model coming straight from AMD.

Here’s the thing, though: when you compare the new cards’ power consumption to their predecessors, you notice that there’s hardly any difference at all. Some of the newer boards technically enjoy higher peak clock rates, though this contributes little to the power story, as our test cases don’t allow for sustained operation at those frequencies.

Let's take a closer look at this situation and the resulting power consumption of these graphics cards:

Clock Rate And Temperature

The same challenges that faced us in trying to create a reference R9 280X for our power consumption testing also affect our frequency monitoring and thermal readings. Again, our U.S. and German offices received different partner boards with a variety of configurations. AMD, for some reason, pinned an embargo on those third-party products that expires later, so we’re setting aside measurements of the 280X for a round-up instead.

What we want to illustrate, though, are the effectively-achievable clock rates under load:

While the R9 280X that we adjusted to behave like a reference model can sustain the higher frequencies in real-world gaming conditions, the R7 260X and the R9 270X behave differently. We got the R7 260X to hit an almost-constant 1100 MHz by using the +20 setting in PowerTune, but saw little difference from the R9 270X. Tuning the card’s settings allows us to hit its peak clock rate more often, but that’s still not enough to call its ceiling a usable increase. This is why we’re using the R7 260X with its two PowerTune setups for our thermal measurements.

Temperatures Under Realistic Load

We took our thermal readings in a closed-up Corsair Obsidian 900D with its case fans spinning slowly and an ambient temperature of 22 degrees Celsius.

The R9 270X’s temperature rises and then levels off at around 80 degrees. In contrast, the R7 260X hits a peak value (with and without our PowerTune adjustment) and then backs off before stabilizing. These differences are the result of an evolved fan controller, which we’ll look at on the next page. Both reference cards seem to be designed for a maximum temperature of 80 degrees Celsius, though. By raising the PowerTune value, we’re able to push that boundary up.

Fan Speed And Noise

Fan Speed

The R9 280X gets left out again, since we want to focus on the reference design, rather than partner boards that’ll all perform dissimilarly.

Since fan speed is one of the primary determinants of noise level, this chart is worth a close examination.

The temperature curves on the previous page suggested what we’d see here. The R9 270X ramps up fan speed in a more granular way, smoothing out the thermal plateau. The R7 260X tries to hold a lower fan speed for longer before stepping up suddenly. That’s where you saw temperatures peak before dropping back down to the 80-degree range.

Noise Level

AMD tends to put its emphasis on partner boards, which it has embargoed for another couple of days. We still wanted to generate some noise data with the reference cards, though.

They both idle under 32 dB(A), and are practically inaudible. Under the load of our custom gaming loop, the R7 260X demonstrates a modest 44.3 dB(A), while the R9 270X at 47.3 dB(A) is notably noisier. We wouldn’t recommend the cooling solution on either reference card. But again, partner boards typically have their own heat sinks and fans that need to be evaluated independently.

Old GPUs Ride Again, But That’s Not A Bad Thing

“What's in a name? That which we call a rose
By any other name would smell as sweet;”

AMD’s GCN architecture debuted almost two years ago in the Tahiti GPU. I’ve mentioned this before, but I paid $550 each for two Radeon HD 7970s to do my own CrossFire testing. And Tahiti lives on today, now at the heart of R9 280X selling for $300. The Pitcairn GPU followed a few months behind in March of 2012. That GPU gets reused today as well in the R9 270X. Bonaire is a much more recent development, emerging in March of this year as Radeon HD 7790 and again today as R7 260X. It’s the only card currently available with AMD’s TrueAudio technology, and it’s the only board in today’s round-up that employs a more sophisticated version of PowerTune able to switch voltage and clock rate very quickly for finer-tuned response to environmental variables.

Unfortunately, when you don’t have higher-performing products to talk about, your only option is to beat up the pricing to attract new buyers. That’s been going on for a while now, and it’s really the reason why R9- and R7-series cards don’t look like great deals compared to the boards they replace. The good news is that AMD’s story doesn’t end today. We still have R9 290 and 290X cards to look forward to, and those are based on new silicon.

There’s another positive in all of this for AMD: in the process of hacking away at its flagship’s price tag, the company pushed Tahiti into a price band Nvidia doesn’t service. True, R9 280X is slower than the Radeon HD 7970 GHz Edition. But the card is also about $30 cheaper. At $300, there’s a $50 premium over GeForce GTX 760, but that board doesn't handle QHD resolutions as well with detail settings cranked up. If I was building a PC to game on a 2560x1440 display and wanted to get in the door as inexpensively as possible without sacrificing graphics quality, the 280X would be my card. That value is why I’ll hand the Tahiti-based board our Smart Buy award. There’s certainly something to be said for revisiting a GPU when it's selling for $200 less than the last time you reviewed it.

I’m not as impressed with the other two cards. R9 270X is a slightly faster Radeon HD 7870 priced as much as $30 more than its predecessor. We know those 7870s won’t last much longer, but for the time being, who’s going to want the 270X at its introductory price? I’m sure Don will let everyone know in his monthly column when the good deals on prior-gen products run out. Until then, the R9 270X is a “meh”, even if it has no trouble blowing past Nvidia's GeForce GTX 660. I'll take a 7870, thank-you-very-much.

The same goes for R7 260X, except this time the pressure is external. AMD is introducing the 260X at the same price as existing 2 GB Radeon HD 7790 boards, which are a bit slower. However, right before launch, Nvidia announced a cut on its 2 GB GeForce GTX 650 Ti Boost down to $150—a $10 premium. For that $10, you get card that's notably faster in a number of benchmarks. Although AMD now offers its TrueAudio technology on R7 260X, developers haven’t done enough with the feature to make it something we can test. Until then, Nvidia has the advantage for its card’s performance.

The R9 290 and 290X will feature TrueAudio, too. Will either of those cards have what it takes to do battle with GeForce GTX Titan or 780? Time will tell, and those are the boards we're most looking forward to.

Want To Know The Back-Story Of GeForce GTX 690 And Titan?

When Nvidia’s GeForce GTX 680 launched, it was both faster and less expensive than AMD’s Radeon HD 7970. But the GK104-based card was still pretty familiar-looking. Despite its conservative power consumption and quiet cooler, the card employed a light, plastic shroud.

That’s not really a big deal in the high-end space. For as far back as I can remember, $500 gaming boards employed just enough metal to support the same plastic shell, but lacked any substantial heft. Cooling fans typically ranged from tolerable under load to “…the heck were they thinking?” And power—well, I was just stoked to see a flagship from Nvidia dip below 200 W.

But then came GeForce GTX 690. And then Titan. And then 780. And then 770. Nvidia’s reference design for each was—and I’m going to sound like a total hardware dweeb—beautiful. Not femininely beautiful à la Monica Bellucci’s lips, but mechanically, like a DBS. The 690 introduced us to a chromium-plated aluminum frame, magnesium alloy fan housing, a vapor chamber heat sink over each GPU covered with a polycarbonate window, and LED lighting on top of the card, controllable through a software applet that companies like EVGA make available to download.

The models that followed actually used slightly different materials for the shroud, and weren’t quite as chiseled. But the aesthetic was similarly high-end. Each time a shipment of cards showed up, I was excited…and that doesn’t happen very often.

It’s Not Easy To Impress An Old-Timer

Alright, at 33 I’m not really that old. But I started reviewing hardware back when I was 18—more than 15 years ago. In the last decade and a half, only a few components really got my blood pumping. Previewing a very early Pentium III and 820-based motherboard with RDRAM months before Intel introduced the platform was cool. So too was the first time I got my hands on ATI’s Rage Fury MAXX. When AMD launched its Athlon 64 FX-51 CPU, it shipped a massive high-end system that I used for months after in the lab. And who could forget the Radeon HD 5870 Eyefinity 6 Edition, if only for the novelty?

I was easily as excited about GeForce GTX 690. I have a dead one in my office, and I swear it needs to live out the rest of its days in a frame on the wall.

Of course, when I’m evaluating a piece of hardware, good looks rank relatively low on the list of attributes that dictate a recommendation. Performance, power, noise, and pricing are also natural considerations. So, in the conclusion for GeForce GTX 690 Review: Testing Nvidia's Sexiest Graphics Card, we pointed out that two GTX 680s are faster and better equipped to handle the heat they generate. Then I brought up then-dismal availability. But all the while, I wanted someone at Nvidia to tell me what it took, at a company level, to make the card’s industrial design a reality when it clearly seemed “overbuilt.” Who was responsible? Would GeForce GTX 690 forever alter the minimum acceptable bling on a high-end gaming card?

At the time, Nvidia still had GeForce GTX Titan, 780, and 770 in its back pocket. When I approached Nvidia’s PR team for a more intimate look at what went into 690, it said “we’ll see.”

That was more than a year ago though, and now Nvidia’s plans are all much more public. Several months back, during a trip up to Santa Clara for a deep-dive on Tegra 4, I sat down with Andrew Bell, vice president of hardware engineering, and Jonah Alben, senior vice president of GPU engineering to talk about the conception of this design that I admired so much and what it took make a reality.

From The Top: Shipping The Hunter Before The Tank

Back in the GeForce GTX 400 and 500 days, Nvidia’s marketing came out with this analogy that resonated with me as a recovered WoW addict. It referred to its highest-end parts leveraging GF110 as “tanks,” meaning they came equipped with all of the compute performance and gaming potential enabled by the Fermi architecture. Later, the GF104-based GeForce GTX 460 surfaced with resources better-suited for playable frame rates, sacrificing most of the bigger chip’s double-precision units in the interest of a leaner GPU. The team used the term “hunter” for that one.

Fast-forward to the Kepler generation. When Nvidia introduced GeForce GTX 680, revealed its GK104 roots, and didn’t discuss compute performance at the launch event, everyone suspected something higher-end was waiting in the wings. As we now know, Nvidia needed around 18,000 GK110s to sell to the government in the form of Tesla cards. So, it made the decision to take GK104 to market first, not necessarily knowing how the card that later became GeForce GTX 680 would fare against AMD’s best effort. In retrospect, its approach paid off. But it could have just as easily gone badly.

Nvidia was working on GK110 all the while, and the processor certainly presented significant technical challenges. The original (65 nm) GT200 GPU at the heart of GeForce GTX 280 measured 576 square millimeters. GK110 covers a still-gargantuan 551 mm², but is also composed of 7.1 billion transistors to GT200’s 1.4. I’ve been told that GK110 is the largest chip that TSMC could fabricate at 28 nm, even. So, looking back, delaying Nvidia’s “tank” in the desktop space allowed the company to do the engineering necessary for a high-performance GPU that behaves well thermally and acoustically.

That’s not to say Nvidia’s decision makers knew this was going to be the case right away. Having just come from Fermi (which you’ll remember drove the infamously-hot GeForce GTX 480), Andrew’s team saw the size of the chip Jonah’s team was working on and immediately assumed that GK110 would be a 300 W GPU. The GPU engineers were confident that Kepler would be more power-efficient though, and, in the end, GeForce GTX Titan with its one disabled SMX hit the 250 W Nvidia’s board guys were hoping for.   

The board team also planned for GeForce GTX 680’s GK104 processor to be a 250 W part, anticipating that it’d replace GeForce GTX 580’s GF110 at 244 W. As we discovered last March, though, the 680 ended up with a 195 W TDP, pushing up to 225 W or so through GPU Boost. But Andrew and Jonah admit to quite a bit of shared stress at the time, as engineers grappled with the previous generation’s power difficulties, what they believed Kepler could do for efficiency, and AMD’s high-profile transition to the powerful GCN architecture, all at the same time.

“The 250 W ID”: Making A Graphics Card Sexy

When you sit down and talk to two people responsible for so much of what a product like GeForce GTX 680, and then all of the other 600- and 700-series boards look like, you can’t help but develop a deep sense of appreciation for what is involved in creating that graphics card. No doubt, the same interplay happens at AMD. But in the case of GeForce GTX 680, Jonah’s team modeled GK104 prior to taping out the photomask for manufacturing, creating an approximate power target. Andrew’s group used that number, added some margin in case the model was off, and built a card capable of accommodating the new GPU.

In fact, the board guys went a step further. Back in the Fermi days, they built one card and made it work at multiple power points. Because add-in board partners take the GPUs and drop them onto their own implementations, that approach was fine. However, this time Nvidia designed three—250, 225, and 200 W—and picked the most complementary to launch GeForce GTX 680 with, which was the middle board. We now associate the 250 W design with GeForce GTX Titan, 780, and 770.

GeForce GTX 690 had different requirements as a result of its two GPUs, though. The centrifugal fan mounted to the back wouldn’t be ideal, for example. So the team created a card and cooling solution for two GK104s.

Originally, GeForce GTX 690 was going to sport the “black coffin” design—Jen-Hsun’s playful nickname for the plastic shroud around GTX 680. But about three months before GTX 690 was to go public, after the engineering had mostly been done and just before production, Jen-Hsun started asking the team if they really wanted to ship the card in a form I have to imagine looked a little like GeForce GTX 590. And that started the team down this thought path of showing off technology, rather than covering it up.

Jonah’s group spends a billion dollars building a GPU, after all. Andrew has a cadre of engineers working on the boards to house those processors. Then they cover them up with plastic. In the past that seemed perfectly natural. But how cool is it to look in the back of a 458 Italia and see that 4.5 L V8 snarling at you through the engine hatch? Wouldn’t you want to peer into your graphics card’s innards the same way? I mean, most boards give you the impression that a company did just enough to keep its technology running within certain design parameters. But GeForce GTX 690 was the first card that made me wonder out loud, “Wow, I wonder how much Nvidia spent dressing this thing up?” As far as the reworked industrial design goes, Andrew’s team came up with something that does do justice to the hardware it hosts.

Andrew calls the idea behind his team’s design the absence of design. Avoid the shaped plastic, avoid stickers, and avoid silkscreen, but expose as much as possible underneath. It’s a design language that the company calls direct and honest, as every detail needed purpose, with no waste on extraneous or trendy characteristics.

GeForce GTX 690 Vs. Titan: The Differences Are Subtle

When you put GeForce GTX 690 and Titan next to each other, the two cards look like they share a great many characteristics, aside from the obvious fan configurations. But because one board dissipates 300 W of power and the other is for a 250 W GPU, there are some differences between them that aren’t readily apparent.

Those subtleties are important to the story. As Andrew’s team sought to “beautify,” they had to be mindful of thermal performance, which suffers if you try to go overboard on the details that simply look nice. The good news was that its dual-GPU (code-named Gemini) configuration indeed fit within that 300 W ceiling. Just one generation prior, GeForce GTX 590 was a 365 W card, requiring a more pointed emphasis on effective cooling. GeForce GTX 690 gave Nvidia a little freedom to get artsy. The cooler is consequently smaller than it would have needed to be for 590.

Here’s an example of how 690 and Titan differ. If you look across the 690’s top surface, you can see the polycarbonate windows recessed slightly, giving the face a greater sense of depth. But the shorter GeForce GTX Titan card, with its 250 W GPU, doesn’t have the same cooling headroom. Its PCB window is flush with the aluminum arms. This allows an additional 4 mm or so on the fin stack, which Andrew says is worth as much as 10% performance from the cooler.

Another example: Nvidia actually designed a back plate to cover the PCB. In the end, the plate was removed because it negatively affected SLI performance. We all know that the PCI Express electromechanical specification is very exact, and there’s only so much room between graphics cards installed back to back. Now, if you look down the top of a GTX 690, the fan sits just a bit lower than the card’s surface. This is done for airflow. Even when two boards are next to each other, that gap is large enough for the top card to suck in the air it needs. Unfortunately, the extra height of a back plate impedes airflow, impacting thermals by anywhere from six to nine degrees, according to Andrew. Were it not for SLI, covering the back of the PCB would improve cooling. So now the engineers are looking for ways to either design a plate that’ll work in SLI or be end-user-removable.

In my reviews, I’ve also commented on the fact that Nvidia went with a magnesium alloy fan housing for GeForce GTX 690, and then ditched it for Titan, 780, and 770. My understanding was that this had something to do with cost. But Andrew told me there really wasn’t much difference between magnesium and aluminum—at least as far as the metals go. More significant was the manufacturing difficulty.

There were actually two GeForce GTX 690 designs, one from magnesium and the other aluminum. The magnesium was a little lighter, though apparently not noticeably so. Its biggest benefit had to do with feature size, and when you set a 690 down next to a Titan, the 690 is just a bit stronger and more prominent-looking. Getting the finish on 690 was a nightmare, though. At first, the goal was to achieve a brushed aluminum look, like what you see from a kitchen appliance. The problem was that putting the shroud’s edges up against a sanding wheel to get the desired effect totally killed the detail. Sharp edges and corners were ground right off.

Once it was determined that brushing wouldn’t work, the team started experimenting with painting, raw finishes, and finally plating. None of it looked the way they wanted. By accident, some previously-plated parts were run through again, yielding an interesting result. And in the end, three plating technologies were combined to yield the finish Nvidia shipped to retail. It actually resembles the oxidized patina finish that some metals acquire over time; the 690 in our lab shows a soft glow of frequently-handled metal, while our Titans perpetually remind us that they’re painted.

But the biggest issue was that manufacturing yields started off around 10%. That’s not the GPU, mind you. Just the coolers. Because they were going from raw magnesium to plating, with no brushing or sandblasting in between, all of the dents from the casting line showed through. Someone had to manually sort through everything coming off the line and pick out the pieces without those imperfections. Even then, if you look closely enough, it’s still possible to pick out tiny divots in the metal.

Nvidia eventually got yields of the 690’s cooler up in the 30-percent range. Nevertheless, it’s understandable that the company quickly switched to a different design. But Andrew is still looking to improve further for generations to come. He wants to build a fan with an integrated hub, for instance, so that when the blades spin, you don’t see the center wobbling at all. It’s the difference between buying a car with hubcaps and one with rims, he jokes.

My Personal Crusade: Acoustics

Beyond the good looks and impressive performance we saw from Nvidia’s GeForce GTX 690 and Titan, I was equally impressed by how unobtrusive the cards were, even under load. If there’s one thing that’ll kill a piece of hardware for me before I start benchmarking, it’s too much noise. AMD’s Radeon HD 6990 was so bad that I felt obligated to record video, lest you guys think I was exaggerating. When the reference Radeon HD 7970 GHz Edition showed up exhibiting acoustic issues of its own, I wondered how such an important characteristic could have gone unnoticed a second time. Finally, when my Radeon HD 7990 showed up, AMD’s team made it a point to mention that the card was designed carefully to operate quietly. And in a single-board configuration, it certainly does.

Nvidia’s record isn't spotless in the quest for quiet computing, though. It has the GeForce FX 5800 Ultra as a reminder of that. But the GeForce GTX 480 seemed to…let’s say encourage a renewed verve for the user experience, and not just frame rates.

You might think that’d be a simple matter of picking the right fan technology, perhaps spending more on a higher-quality blower, dialing in a smooth ramp, and hoping the engineers who designed your GPU kept efficiency in mind. But it’s not; there are trickier scientific forces at play. A few years ago, the board team decided to do something about squealing power circuitry and loud fans, kicking off two distinct efforts: component acoustics and fan acoustics.

The former required specific hires and specialized equipment for analyzing how the board and its on-board components move. Of course, when Nvidia chose to go down this path, it didn't know where the noises were coming from. The engineers figured out that there were two sources. First, they identified a ton of inductors that were put together poorly. As their glue let go over time, you’d end up with squealing. But even with good inductors, the board itself resonates. The SLI connector, for example, flaps around. At a certain point, that frequency ends up in an audible range and we hear it. So there are folks responsible for screening each product that gets built to catch those noises.

Interestingly, most of the prototypes that come back do make noise. When a power supply responds to a GPU idling at 25 W and jumping up to 250 W, a lot of those on-board components start moving, causing the board to flex, which you sometimes hear. Sometimes a few pieces can be relocated to fix this, but other times the whole layout has to change.

Optimizing Fan Technology For Less Noise

Then there’s the fan. Several different variables affect its performance. First, how fast is it spinning? The GeForce GTX 480 was a bit of a learning experience for Nvidia’s engineers. Just take a look at GeForce GTX 480 And 470: From Fermi And GF100 To Actual Cards! if you don’t remember how that played out. Without rehashing too much painful history, the board was both hot and loud. In the wake of the 480’s launch, Jen-Hsun and several other employees sat down and listened to various cards at different noise levels for weeks. In the process, Nvidia’s engineers settled on a 36 dB specification that later became the target ceiling for GeForce GTX 680.

Of course, getting there required an effective cooler, which is why product launches are often accompanied by talk of embedded heat pipe designs and, even more premium, vapor chambers that help cut temperatures by an additional eight to 12 degrees. If you assume five watts per degree Celsius, that’s up to 40 to 60 W of additional cooling at the same acoustic level. Take two competing GPUs rated for similar power, then, arm one with the more effective heat sink and deprive the other of comparable thermal performance. This comparison goes a long way in explaining my critique of acoustics in AMD Radeon HD 7970 GHz Edition Review: Give Me Back That Crown! Nvidia had to learn its lesson with GTX 480, which paved the way for a much less obtrusive 580. And the culmination of that work, which continued evolving after Fermi, is Titan’s sink.

Second, how are the fan blades and airflow interacting within the cooler? A cooling solution can be not loud, but still annoying. Even at idle, we’ve had cards in the lab that exhibited a high-pitched buzz or whooshing of air. The buzz often comes from the motor. And for Nvidia, avoiding it meant working closely with fan vendors on a spec that defined what the motor needed to do. Suppressing the whistling noise requires some optimization of the scroll at the bottom of the fan, and again, this is something that had to be researched, identified, and then engineered around. To the layperson this all sounds easy, but if it were, we wouldn’t feel so strongly about the discrepancy between some of the offending cards we've reviewed from both AMD and Nvidia.

Despite his team’s progress, Nvidia’s Bell says he’s not completely happy with Titan’s acoustics and wants to cut down on motor noise even more. The experience he’s going for is getting into your car and turning on the engine. There’s a massive fan in there that’s really loud as it pours air across the components under your hood. But it’s not an issue in the cabin, and the noise you do hear isn’t offensive. Of course, if you have a GeForce GTX Titan you also know that, while it ships with a conservative fan ramp, it can also be tuned more aggressively if you’re willing to tolerate more noise.

Building Better Boards

And so we come full circle. I thoroughly enjoyed sitting in a room with the guy who designs Nvidia’s graphics cards and the guy responsible for the GPUs on those cards. I don’t think graphics performance came up once during our discussion. Rather, we had plenty to talk about by just focusing on how both teams are making the high-end graphics experience more special. Particularly after ribbing Intel a bit in Intel's 12-Core Xeon With 30 MB Of L3: The New Mac Pro's CPU? for the lack of innovation happening in desktop processing, Nvidia shows us that it’s able to build a significantly faster graphics architecture that uses less power than its predecessor, is more acoustically-friendly, and is wrapped up in materials like aluminum, magnesium, and polycarbonate.

Yes, the GeForce GTX Titan is very expensive. I’ve said before that there are only two places I’d recommend it for $1000: in a gaming-oriented small form factor chassis, as the fastest single-GPU card you can buy, or in a multi-card configuration where money is no object. But even though its endorsement is neither universal nor emphatic, I’m still very impressed by GK110 and Nvidia’s industrial design. The GeForce GTX 780 was almost equally impressive, though I think that anyone who can afford a fast $600 card would prefer the faster $1000 option instead.

But don’t forget that it was the GeForce GTX 690 that first wowed us back in April of last year. Going from this…

…to this made the sort of impression on me that only ATI’s Rage Fury MAXX, AMD’s FX-51, and the occasional super-early engineering sample have matched.

This stuff is the product of passion, and it’s how we know the enthusiast market is still going strong, despite prognoses that the PC in general is waning. And just look at what Nvidia’s doing next: taking that same Kepler architecture and leveraging its composition to drive the graphics component in its next-generation SoCs.

AMD has confirmed that it has hired on Raja Koduri, a former director of graphics architecture at Apple. He will be rejoining AMD after leaving the company four years ago, and will assume a role in AMD's graphics business as corporate vice president, Visual Computing. He will be reporting to Mark Papermaster.

“Maintaining AMD’s position as a leader in visual computing is the key to our long-term success. As one of the industry’s foremost experts in developing leading-edge visual computing solutions, Raja brings exceptional vision and strength to AMD’s world-class engineering leadership team,” said Papermaster. “Given his past record of success, Raja will help AMD lead the way in visual and accelerated computing and ensure we continue developing the industry-leading graphics IP that forms the foundation for our future growth.”

Koduri began his career serving as Director and Engineering Manager at S3 Inc., and then moved on to ATI Technologies as the Director of Advanced Technology Development in 2001. Naturally he was assimilated into the AMD collective when the company acquired ATI in 2006, and eventually moved up the corporate ladder as the chip giant's Chief Technology Officer of the Graphics Product Group before leaving in 2009.

Apple reportedly hired Koduri as part of the company’s push to build up its own chip operations. Rather than rely on partners like Intel to design SoC's for its tablets and smartphones, the company began creating its own chips.

After departing Apple earlier this year, Koduri served as Technical Advisor to the Board of Directors at Makuta VFX. This company is described as "a fully-fledged visual effects facility covering a full gamut of requirements from active on-set visual effects supervision through to immersive digital set extension, digital matte painting, high-end feature animation and effects work, clean-up, motion tracking and final compositing."

AMD and Apple have seemingly been tossing executives back and forth over the past few years. In August 2012, AMD named Jim Keller as its Corporate Vice President and Chief Microprocessor Architect. He previously served as a director in Apple's platform architecture group, and even worked for AMD prior to that. Keller now reports to former Apple hardware chief Mark Papermaster. The previous hardware chief was John Bruno who left AMD for Apple.

According to the report, AMD had courted Koduri for some time, wanting him to return to the company. Koduri and Keller efforts at Apple complimented each other, and will likely continue with both under the same AMD roof. Their expertise will reportedly help accelerate AMD's System-On-A-Chip strategy of cramming all the features it can onto a single central chip.

AMD reported a narrower first-quarter loss on Thursday, but its revenue dropped a huge 31-percent in a year-over-year comparison. To combat the slump in sales, the company has laid off a large portion of its staff and restructured its relationship with Globalfoundries. However in its report the company highlighted the announcement of Sony's APU-powered PlayStation 4, the Never Settle: Reloaded program, Radeon Sky Graphics, among others.

Is AMD Self-Loathing?

For years, we heard that ATI's graphics cards are more platform-dependent than Nvidia's and, depending on who had the fastest processor at the time, should really be used with that CPU. So, when AMD's highest-end processors started falling further and further behind Intel's quickest models, we weren't surprised when Nvidia started introducing AMD-compatible chipsets. Intel even forged a similar partnership with ATI, and we looked forward to the RD600 platform overshadowing Intel's own 975X as the premiere enthusiast chipset for Conroe-based processors. 

Many of us were confused when AMD decided to buy ATI rather than solidify its ties to Nvidia. Intel abandoned ATI's RD600 altogether and went off to develop X38 Express. Nvidia eventually dropped out of the PC chipset business entirely. But enthusiasts still took comfort in the notion that AMD’s acquisition might carry it through the rough times ahead. ATI was, after all, slightly more competitive.

Now that AMD and ATI are integrated (as well as two large companies can be after several years), we'd expect its CPU and GPU technologies to be extensively optimized for each other. Nevertheless, the suggestions continue that Radeon cards need more processing power behind them to achieve their performance potential. If that's true, the implication is that whenever one of our Intel-based platforms shows a Radeon and GeForce card performing similarly, an AMD-based system would actually show the GeForce performing better. Wait. What?

We began our tests with an evaluation of clock rate and its effect on CrossFire in FX Vs. Core i7: Exploring CPU Bottlenecks And AMD CrossFire. Intel started out at a lower frequency and consequently had the most to gain. AMD couldn’t go very far beyond its stock clock rate without more exotic cooling, so it had the least to gain.

At the end of the day, both of our CPUs ended up at comparable clock rates with similarly-little effort, making that article a great head-to-head match. But that slight speed-up from AMD meant that a second GeForce-based article with the same CPU settings wouldn’t have given us very much new information. So, I decided to jump straight to the point: Does AMD’s flagship FX processor, overclocked, favor Nvidia graphics?

Test Settings And Benchmarks

Great performance and quick installation keep Thermalright’s MUX-120 and Sunbeamtech’s Core Contact Freezer in my inventory of favorite testing components. The brackets that come with these older samples make them non-interchangeable, however.

G.Skill’s F3-17600CL9Q-16GBXLD has a remarkable DDR3-2200 CAS 9 rating, using Intel XMP technology for semi-automatic configuration. As a non-Intel platform, the Sabertooth 990FX configures XMP values through Asus' DOCP setting.

Seasonic’s X760 provides the consistent efficiency required to assess platform power differences.

Keeping the benchmark set from our previous round cut back testing time, though it also meant utilizing older drivers. The thing to remember is that we aren't trying to compare the performance of AMD's and Nvidia's graphics cards, and we're breaking each GPU vendor into separate charts to prevent this. Rather, we're interested in how each configuration behaves attached to AMD- and Intel-based platforms.

Results: Aliens Vs. Predator

Again, we're making every attempt not to create a Radeon or GeForce graphics card review; the focus of this article is more specific. In order to address murmurs about AMD's platform dependencies, I want to know how each vendor's hardware scales on AMD’s flagship CPU. Two charts represent two different graphics sources, and ideal scaling would show two charts that look identical, apart from their actual average frame rates.

Our less-demanding Aliens vs. Predator results demonstrate the ideal CPU-to-GPU scaling scenario. Even though AMD's Radeon HD 7970s are faster, the charts are a near-perfect match, suggesting similar scaling.

We again see consistency between Radeon and GeForce results. AvP prefers the Radeon cards, but when we ignore absolute frame rates, the bars appear nearly identical. If this remains true through all of our testing, we can fairly confidently "bust the myth" that Radeon cards require higher-performance platforms to reach their potential, and that AMD's FX CPUs bottleneck Nvidia's graphics products less.

Results: Battlefield 3

The Radeon graphics cards encounter a weird performance penalty at 1920x1080, regardless of the CrossFire mode or processor used. The consistency we're looking for doesn't come from the first and second charts, but rather from comparing the top half to the bottom half of each data set. Doing so tells us that the FX-8350’s small performance deficit doesn't come from the graphics cards we're using, but rather the processor's performance.

Incidentally, comparisons to the results from our most recent System Builder Marathon reveal that newer drivers ameliorate the anomaly at 1920x1080.

Results: F1 2012

We’ve already seen that F1 2012 is bottlenecked by memory throughput, so we’re not surprised to find AMD’s less effective memory controller holding back average frame rates at the low to medium settings that emphasize platform (rather than GPU) performance.

The only consistency we find appears to be a bottleneck outside of the graphics system, so we looked closer at the differences between single- and multi-GPU configurations.

Adding a second card to the FX-8350 platform actually drops F1 2012’s performance at its High Quality preset, with CrossFire imparting a bigger penalty than SLI. This game's Ultra detail setting is necessary to push the graphics workload beyond CPU and memory bottlenecks.

The FX-8350 appears to be a big limitation to Radeon HD 7970 performance. What happens when we switch to GeForce?

Looking exclusively at our 5760x1080 results, a single Radeon HD 7970 appears far more powerful than one GeForce GTX 680 in this game. We also see that the same GeForce GTX 680 appears far less CPU-bound than a single Radeon HD 7970 when we match either card up to AMD's FX-8350 processor. Unintended (and not ideal) though it might be, AMD's CPUs might really be allowing Nvidia's graphics cards to come closer to their potential than its own.

The FX-8350 appears to completely bind up CrossFire performance, while allowing SLI a little more space to stretch its legs at higher resolutions. This is again evidence of a platform-oriented issue, at least in one game thus far.

Results: The Elder Scrolls V: Skyrim

The FX-8350 appears to hamper the frame rates of CrossFire and SLI configurations alike in Skyrim. More taxing detail levels have little effect on the shape of the chart, though the numbers assigned to those colored bars are a little lower.

Results: Metro 2033

CrossFire performance scaling is tighter than SLI scaling in Metro 2033. Although this applies to both Intel and AMD CPUs, only the FX-8350 is able to pull down a pair of Radeon HD 7970s to the frame rates a single Radeon HD 7970 achieves. That’s a shame, since the company’s Tahiti graphics processor appears more powerful than the GeForce GTX 680's GK104.

Frame Rate-Over-Time Analysis

Our first article included a frame rate-over-time analysis intended to identify problematic sequences in our testing. Regardless of whether you're using one card or multiple GPUs, dramatic slow-downs interrupt game play. Unfortunately, although Nvidia enabled Don (up in Canada) and Chris (down in Southern California) with its FCAT tools, I'm only able to use the Fraps-based testing I ran previously. We know from our comparisons in Challenging FPS: Testing SLI And CrossFire Using Video Capture that Fraps isn't able to accurately capture the dropped and runt frames that might plague one graphics solution but not the other. However, we are at least able to track when each combination of cards drops to levels we deem unplayable.

Our least-demanding Battlefield 3 settings should reveal CPU bottlenecks, and indeed we see that three out of four multi-GPU configurations turn in similar results. Only Intel's Core i7 is powerful enough to butt up against the game’s 200 FPS cap, and only when it’s paired with two GeForce GTX 680s in SLI.

Our most taxing settings should demonstrate GPU limits. Unfortunately, those limits appear fairly consistent for all four dual-GPU configurations. At least the test runs smoothly, staying well above 50 FPS throughout its duration.

An attempt to demonstrate CPU-bound conditions in Skyrim is somewhat successful, with noticeable separation between the Intel and AMD CPU results. Both processors favor SLI over CrossFire in this title.

And now for our most taxing detail settings in Skyrim, where graphics performance matters most. Two Radeon HD 7970s in CrossFire beat the GeForce GTX 680s in SLI on the Intel-based system, while the reverse holds true for AMD's FX-8350. Doh!

Results: 3DMark 11

We've seen 3DMark favor Nvidia's graphics cards and Intel's CPUs. So, we're adding these numbers for the folks who track results across multiple sites, but we're not using them in our overall performance evaluation.

3DMark's Extreme detail preset is constrained a little more by graphics performance, reducing Intel’s margin of victory.

Power And Efficiency

Despite the marketing behind ZeroCore, and indeed, the technology suite's effectiveness in single- and multi-card configurations, Radeon HD 7970s cannot idle with three monitors attached. The host processor's power use isn't bad, though. This is one of those instances where putting the AMD and Nvidia cards into separate charts make sense, since we're trying to compare CPU-to-GPU pairing, rather then CPUs or GPUs alone.

Regardless of whether you're running under an AMD or Intel processor, adding a second Radeon HD 7970 appears to impart far greater power consumption than a second GeForce GTX 680. Single-GPU load power is comparable between the competing graphics cards.

Those power draw differences are reflected in reduced efficiency. Comparing performance to power, GeForce efficiency appears to increase in SLI, while Radeon efficiency appears to drop in CrossFire. None of this gets us closer to figuring out whether AMD’s fastest CPUs allow Nvidia's graphics hardware to reach further than its own, however.

CPU-To-GPU Performance Scaling

Much to the chagrin of my boss, I followed up a week of benchmarking with a few days of wading through data, trying to figure out the best way to present these results. We can, for example, consider the cards first.

We clearly see that the Core i7-3770K gets a bigger boost from two Radeon HD 7970s in CrossFire than AMD's FX-8350, but that could just be a general indication of lower CPU performance from the Vishera-based CPU. That is fair enough. The FX is a less expensive processor, so we're completely fine with it not performing as well.

SLI scales a little better than CrossFire on the Intel CPU, and the SLI-versus-CrossFire spread is even wider on the FX-8350-based configuration. We’re getting close to an answer.

Using the Core i7-3770K as a common factor, we find that AMD's Radeon HD 7970 slightly outperforms the GeForce GTX 680 in our test suite. Yet, slightly better scaling allows SLI to catch up to CrossFire.

Conversely, the FX-8350 appears to favor Nvidia's SLI technology over CrossFire. This is just another data point quantifying the potential validity of claims that AMD's cards are more limited on its own processors, and that Nvidia's graphics hardware is able to extract more performance from a top-end FX.

Intel takes an 11% lead over AMD when paired to a single Radeon HD 7970. That lead shrinks to only 9% when a single GeForce GTX 680 is used. Of course, this chart doesn’t make clear whether AMD is favoring the GeForce, or Intel is instead favoring the Radeon.

CrossFire boosts gaming performance on Intel's Core i7-3770K by 72%, but only manages to speed up frame rates on AMD's FX-8350 by 47%. The Vishera-based FX gets a far bigger kick from SLI.

How Does FX Treat Your Graphics Card?

Does AMD’s flagship CPU really allow GeForce cards to run closer to their potential than Radeons? That conclusion is still a little difficult to reach, since the Radeon HD 7970 is faster than the GeForce GTX 680 in today's benchmark suite. We can still get there by comparing the results of a less-bottlenecked processor, however, and we can get even more information by comparing CrossFire to SLI. After spending way too much time poring over the data, the best illustration of CPU-to-GPU scaling occurred as I wrote this article’s introduction.

The above chart shows that, without significant CPU bottlenecks, two GeForce GTX 680s in SLI and a pair of Radeon HD 7970s in CrossFire offer similar performance. It also illustrates that this performance similarly comes not from having identically-performing cards, but from a slight scaling advantage favoring the slower GeForce GTX 680s when you put two of them together.

Our big revelation is that AMD’s FX-8350 performs 5% better in SLI than it does in CrossFire. When all else is made equal, AMD’s current flagship host processor really does favor Nvidia's graphics technology. Whoops.

Welcome To AMD's Campus In Markham, Ontario

Located just north of Toronto, AMD keeps the fire lit in its campus' main atrium. On October 20 and 21, the company hosted ExtravaLANza, a LAN party for enthusiasts and an open exhibition area where partners like Ubisoft, Asus, MSI, PowerColor, Diamond, Sapphire, XFX, Gigabyte, Newegg Canada, NCIX, Cooler Master, NEC and SilverStone all had things on display.

ATI Mobility Radeon Museum

On display in the halls of AMD's meeting rooms are GPUs chronicling the company's past as ATI. Obviously, this collection covers previous-generation Mobility Radeon chips.

BYOC Area

This is the BYOC area shortly after the Team Fortress 2 tournament started.

Gamers Gaming

Here's a group of serious gamers, each armed with his or her choice of headphones.

Kickin' It Old School

Some guys from Oakville, ON brought CRT-based monitors.

Check Out Those Tubes

Here's one gamer's machine. A see-through 5.25" bay gives you a great view of the braided tubing inside.

Gamers' Builds

Most of the attendees were proud to show off their rigs, and with good reason.

A Clean Job

We were impressed by this exceptionally clean and tidy configuration.

Lara's Back

Moving back out into AMD's main atrium, we find a massive banner advertising the upcoming Tomb Raider game.

AMD's Doors Its Open To The Public

The event got fairly busy once AMD opened its doors to the public. Hardware vendors were giving out promotional items and bigger prizes.

Family Fun In 3D

Some families even made the event their weekend entertainment.

I Have To Get Me One Of These

AMD's 5x1 Eyefinity configuration never fails to impress, and it's even better from a force-feedback racing seat.

Cooler Master Goes Gothic

This custom Cooler Master chassis would fit in at any Halloween party.

Ruby Rocks A Case

ATI's infamous Ruby character graces a custom chassis from NCIX.

Sapphire Goes Mini With An APU

Sapphire was showing off an upcoming miniature PC that'll feature an AMD A8 APU. The final version won't have a clear case, but so many attendees liked it that we wouldn't be surprised if Sapphire considered it as an option.

Hands-On With Far Cry 3

We got to go hands-on with Far Cry 3. Not surprisingly, it features more outdoor action.

Playing Hitman: Absolution

Hitman: Absolution is also optimized for Eyefinity-based display configs. The game looked like quite a bit of fun from our limited time with it.

Build Your Own PC

We also visited the builder's section in the BYOC room. A handful of AMD employees were hanging out, putting their own machines together.

Three-Way CrossFire

We're pretty sure this can play Crysis.

MSI's Upcoming GX60 Notebook

We got an early peek at MSI's GX60 notebook, packing a 2.3 GHz, quad-core A10-4600M and a Radeon HD 7970M. It's quite a beefy gaming laptop.

Gaming Evolved

AMD also briefed us on its Gaming Evolved initiative that it hopes will spread support for some of its key technologies, but without locking out competing features, which is something we've seen from Nvidia in PhysX and CUDA, for example.

A Long History With Tom's Hardware

We spotted these awards on the shelves at AMD. They're from almost a decade ago!

The Story Of Fusion Begins

“Nothing is more difficult than the art of maneuver. What is difficult about maneuver is to make the devious route the most direct and to turn misfortune to advantage.”

    —Sun Tzu, The Art of War

When I interviewed Dave Orton, then the president of ATI Technologies, in 2002, one of the first things he told me, "It’s always what’s possible in the business that keeps people going." No more prophetic words could describe the coming merger between his company and CPU manufacturer AMD. The big question, of course, isn’t what is possible or even whether the possible will become reality. The real question is whether the possible will become reality soon enough.

Orton spent most of the '90s with Silicon Graphics and, in 1999, when almost anything in technology seemed possible, he left SGI to join a little core logic startup called ArtX. The little company won the development contract for Nintendo’s GameCube, which went on to sell a few units (somewhere north of 20 million). That fall, ArtX showed its first integrated chipset at Comdex, and immediately the company flashed on the industry’s radar as a prime acquisition target.

Ultimately, ATI was able to put ArtX in its pocket and made Orton its president and COO. Then the tech bubble burst, driver problems abounded, schedules slipped, and, for a while, it seemed that ATI could do nothing right.

Part of the road back to glory hinged on Orton figuring out how to complete the meshing of these two development teams. He was the one who figured out how to get ATI on a 12-month cycle for new architectures and six- to nine-month cycles for iterative design revisions. Product teams were given more control and responsibility. And slowly, over 18 months, perhaps, with Nvidia kicking it in the ribs at every turn, ATI managed to get back on its feet. The company rediscovered how to execute.

"Just step back and understand your roots," said Orton. "Constantly build. You can never be satisfied with where you are. You’ve got to be satisfied with where you can be and then drive to that."

Back on top of its game, Orton knew it was time to keep driving—but to where? I detected no glimmer of the future in our 2002 discussion. ATI continued to excel at integrating graphics into northbridge chips, and Intel, which still viewed integrated graphics as only needing to be good enough for business apps, was still more of a partner than a competitor.

However, in a keenly prescient moment, Orton told me, "I guess if I could change one thing about computing, I’d like it to be more open to create a broader range of innovation. I recognize the advantages of standards. Standards provide opportunity."

At two different points in our conversation, Orton lamented his daily Silicon Valley commute, even saying that if he could invent anything, no matter how fantastic, it would be a Star Trek-esque transporter. So perhaps we can take him at his word when, in 2007, he left his post as executive vice president of AMD in order to spend more time with his family. But this is jumping ahead. First, Orton’s drive from Toronto was about to take a hard southern turn, straight down to Texas.

Looking For The Other Half

Right around the time I was speaking with Dave Orton, AMD founder Jerry Sanders was stepping down to begin his well-earned retirement. Since just after its founding in 1969, Sanders had led AMD through over three decades of highs and lows, cementing its place as the only major rival to Intel in the global CPU market. Sanders hired Hector Ruiz away from Motorola Semiconductor in 2000 to become his right-hand man (as president and COO) and next in line for the CEO's chair. Two years later, Sanders took his final bow and Ruiz seized AMD's steering wheel.

Meanwhile, Dirk Meyer, a former processor engineer for DEC and Intel, was rising through AMD's ranks. Meyer led the team in 1998 and '99 that produced the Athlon (K7), a design so successful (and based on a bus derived from DEC) that AMD crushed the then-leading Pentium III and beat Intel to the 1 GHz threshold.

In 2003, AMD followed up its Athlon success with the K8 (Hammer) architecture, which almost immediately trounced everything Intel had going in the server market. Intel’s NetBurst, far from realizing its 10 GHz aspirations, turned out to be a blistering disappointment as Intel slowly began to realize that efficiency trumps brute force. Intel didn’t have a follow-up ready until 2006, when it released its Core architecture, which in turn presaged the groundbreaking Nehalem design of 2008.

We all know that the Fates are fickle in technology. Everything ebbs and flows. If Athlon lit the fuse under AMD, Hammer took the company into orbit. By 2005, Ruiz sensed it was time to try and outflank his bigger competitor. That December, Ruiz elevated Dirk Meyer to chief operating officer of AMD’s microprocessor business, effectively making Meyer number two in the company. By that time, the two of them understood that having graphics on the northbridge was good...but not good enough.

According to Forbes, AMD approached Nvidia about merging. As we go forward in this story, consider the element of personalities and corporate cultures, and ask yourself if AMD could have survived such a mix. The answer is likely found in Nvidia’s response: CEO Jen-Hsun Huang was willing to entertain the idea, provided that he would be the chief executive of the resulting organization. Ruiz, at that point feeling on top of the microprocessor world, understandably went looking elsewhere.

In July of 2006, AMD announced that it would buy ATI for the princely, almost hyperbolic, sum of $5.4 billion at a time when AMD was worth somewhere shy of $9 billion. According to Joe Macri, who was then ATI’s director of engineering and another Silicon Graphics alumnus, it was a "grand vision" spawned from Dave Orton and Dirk Meyer.

"There was a lot of risk on AMD’s part," says Macri, now as AMD’s product chief technology officer. "But there was a lot of courage between Dirk and Dave. They could see a future of the need to converge the CPU and GPU in a way that would allow it to be treated as a unified compute model. That initial vision sounded simple. It started with a big business deal that was quite the effort to pull off. And then they brought us together as leaders and said, ‘Do it!’"

"Quite the effort" is an understatement of epic proportions.

Merger And Mayhem

In the summer of 2006, most people didn’t grasp the strategy of what would eventually be called Fusion, the melding of CPU and GPU on a common die. Like most, Ars Technica at the time viewed the merger as a way for AMD to bolster its portfolio breadth, freeing it from reliance on third parties for chipsets and expanding the company into areas such as ultramobile graphics and digital TV.

For their part, AMD and ATI remained thoroughly mute. While some of the silence may have been mandated, owing to lengthy legal process of two large companies melding, a more practical reason might have been the safeguarding of existing sales.

"At any point in time, you’re married to a lot of partners," says AMD’s Macri. "I use the word married because they’re very deep relationships, both in business and, at the end of the day, a personal level. Business is about personal relationships. We make commitments to each other. We might embody them in contracts, but part of that is we’re making a big personal commitment. And you are only as good as your word. If you throw the grand vision out there without the time that it takes to move all your partners to the vision, you lose all your partners. AMD at the time had Nvidia as a very strong chipset and graphics partner. You can’t flip those relationships on and off like a switch. So the guys were somewhat limited in their ability to explain to the world this grand vision and how it would all play out."

In early 2006, AMD’s stock price hovered just above $40 per share. One year later, at a time when the market was nearing its pre-recession peak, AMD had tumbled to under $15. Two years, later, it was bouncing on a $2 floor. A five-year comparison between AMD and Intel shows the story from pre- to post-recession. While Intel looks relatively flat, the rise and fall of AMD is as exhilarating as it is heartbreaking.

Economic downturn aside, what happened? Heading into late 2006, AMD entered into the first of what would become seven consecutive quarterly losses preceding Hector Ruiz’s resignation. Intel’s Core architecture was out and ramping. Nvidia’s GeForce 7-series, launched in June 2005 to considerable fanfare, gave way to the even better 8-series in November 2006. Meanwhile, the delay-plagued ATI Radeon X1000 series arrived in 2005; there was no major 2006 update. The follow-up Radeon HD 2000 didn’t launch until April 2007—in Tunisia, of all places—and even though AMD/ATI’s performance was starting to edge back up, its momentum in the market had significantly slipped.

And those were just the visible problems. Behind the scenes, in the back rooms where the two companies were trying to figure out how to coexist and blend, matters were even more muddled.

Scaling The Brick Wall

This AMD core team found itself with two fundamental problems, one technical and the other philosophical, and both had to be solved before anything could move forward.

"On a pure technology and transistor side, we had a conundrum on our hands," says AMD’s Macri. "What makes CPUs go really fast ends up burning a lot of power on a GPU. What makes GPUs go really fast will actually slow a CPU down incredibly. So the first thing we ran into was just getting them to live on the die together. We had the high-speed transistor combined with the very low-resistance metal stack that’s optimal for CPUs versus the GPU’s more moderate-speed transistor optimized around very dense metalization. If you look at the GPU’s metal stack, it looks like the letter T. It looks like the letter Z in a CPU. One’s low-resistance, one’s lower density, and so higher resistance. We knew we had to get these guys to live on the same die where they both perform very well, because no one’s going to give us any accolades if the CPU drops off or the GPU power goes up or performance falls. We needed to do both well. We very quickly discovered that wall."

Imagine the pressure on that team. With billions of dollars and the company’s future at stake, the group eventually realized that a hybrid solution couldn’t exist on the current 45 nm process. Ultimately, 45 nm was too optimized for CPU. Understanding that, the question then became how to tune 32 nm silicon-on-insulator (SOI) so that it would effectively play both sides of the fence. Of course, 32 nm didn’t exist outside of the lab yet, and much of what finally defined the 32 nm node for AMD grew from the Fusion pursuit.

Unfortunately, until the 32 nm challenge was solved, Fusion was at a standstill—and it took a year of work to reach that solution. Only then could design work begin.

Meanwhile, the Fusion team was also fighting a philosophical battle. With the transistor and process struggle, it was massive, but at least the team knew where it needed to go and what the finish line looked like. Even with the transistor challenge figured out, the question still remained of how to best architect an APU.

"One view was like, the GPU should be mostly used for visualization. We should really keep the compute on the CPU side," says Macri. "A second view said, no, we’ve gotta split the views across the two halves. We’ve got this beautiful compute engine on the GPU side; we need to take advantage of it. There were two camps. One said things should be more tightly coupled between the CPU and GPU. Another camp said things should be more loosely coupled. So we had to have this philosophical debate of deciding what we should treat as a compute engine. Through a lot of modeling, we proved that there was an enormous advantage to a vector engine when you have inherent parallelism in your code."

This might have seemed obvious from ATI’s prior work with Stream, but the question was how much work to throw at the GPU. Despite being highly parallel, GPUs remain optimized for visualization. They can process traditional parallel compute tasks, but this introduces more overhead. With more overhead comes more impact on visualization. With infinite available transistors on the die, one could just keep throwing resources at the problem. But, of course, there are only a few hundred million transistors to go around.

"Think of all the applications of the world as a bathtub," says Macri. "If you look at the left edge of the bathtub, we call those applications the least parallel, the ones with the least amount of inherent parallelism. A good example of that would be pointer chasing, right? You need a reference. You need to go grab that memory to figure out the next memory you gotta go grab. No parallelism there at all. The only way to parallelize is to start guessing–prediction. Then, if you go to the right edge of the bathtub, matrix multiply is a great example of a super-parallel piece of code. Everything is disambiguated very nicely, read and write stream is all separate, it’s just beautiful. You can parallelize that out the wazoo. For those applications, it’s very low overhead to go and map that into a GPU. To do the left side well, though, means building a low-latency memory system, and that would load all kinds of problems into a GPU that really wants a high-bandwidth, throughput-optimized memory system. So we said, 'How do we shrink the edges of the bathtub?' Because, the closer we could bring those edges, the more programs we could address in a very efficient way."

A big part of the philosophical debate boiled down to how much to shrink those bathtub edges while preserving all of AMD’s existing visualization performance. Naturally, though, while all of this debate was happening, AMD was getting hammered in the market.

Up From The Ashes

No one on the outside could see the engineers frantically fighting for answers and a path forward. What they saw were stony-faced executives and delays. Lots and lots of delays in multiple segments from graphics to CPUs to chipsets. In July, ten months after the merger, AMD executive vice president Dave Orton, arguably one of the most influential minds behind today’s hybrid processor trend, resigned. In September, chief sales and marketing officer Henri Richard followed Orton out the door.

Ultimately, the $5.4 billion decision to buy ATI rested on Hector Ruiz’s shoulders, and onlookers couldn’t help but associate this purchase with the catastrophic plummet in AMD’s financials.

Sources at TheStreet.com indicated that Ruiz might also be resigning, although his contract ran until April 26, 2008. As it turned out, Ruiz survived another three months, finally leaving in July.

Ruiz went on to become the first CEO of GlobalFoundries in March 2009. More significantly, GlobalFoundries started out as AMD’s spun-off manufacturing arm. With its various setbacks, AMD could no longer afford to maintain so much manufacturing capacity all for itself. This marks the latest and hopefully last major divestiture of the company’s assets, capping 2008’s jettisoning of the digital TV business and (this one really had to hurt) early 2009’s $65 million sale to Qualcomm of the old ATI handset division and all of the mobile graphics and multimedia intellectual property that went with it. By that point, AMD had already written off $3.2 billion in bottom-line value.

Of course, as with most major downturns in life, so long as you don’t stop moving, you’re not dead—and the odds are that things will improve. Having shed much of its former self, AMD is now left as a predominantly R&D- and IP-based company. It’s smaller, lighter, and more flexible. But is that enough? AMD’s board didn’t seem to think so. In January 2011, the last of the old directorate, Dirk Meyer, was pushed aside, according to the press release, to "accelerate the company’s ability" to "have significant growth, establish market leadership and generate superior financial returns."

Some argued that this was undeserved for a guy who apparently brought AMD back from the brink of ruin and saw the first fruits of his Fusion "grand vision" finally start to reach the market. But others questioned at the time of Meyers’s ascendance to the top post whether he had the sales chops to make the big deals that AMD so desperately needed.

With that in mind, one might examine his replacement, Rory Read. Read spent 23 years at IBM, where he held a "broad range of management positions." Following IBM’s PC division, Read then moved to Lenovo in 2006 and eventually became its president and COO in 2009. During his time there, Lenovo became the third-largest PC vendor in the world. Read was appointed president and CEO of AMD on August 25, 2011. Not even a year into the job, it’s still too early to render a verdict on Read. But if the AMD board was hoping for a guy that would yield big deals and new directions, it appears they got exactly what they wanted.

Fusion Ignites

Throughout this shuffling of top office name plates, AMD engineers continued their dogged pursuit of Fusion. What began as a team of four people—former ATI vet Joe Macri, the recently deceased AMD fellow Chuck Moore, then-graphics CTO Eric Demmers (now at Qualcomm), and AMD fellow Phil Rogers, who was the group’s technical lead—had grown to envelop the top three layers of engineers from both the CPU and GPU sides of the company. Macri describes the early phase of their collaboration as "the funnest five months I’ve ever had." The first 90% of the Fusion effort was an executive engineer’s dream. "The last 10% was excruciating pain in some ways."

"That effort resulted in a couple of things," adds Macri. "One, we ended up with the best architecture out there that’s unifying scalar and vector compute. It blows away what [Intel] did with Larrabee. The Nvidia guys have only attacked part of the problem, because they only have the IP portfolio to attack part of the problem. What they’ve done isn’t bad. It’s actually good for having one hand tied behind their back. But with [Fusion], we had the full IP capability, and it truly is the first unified architecture, top to bottom."

Technical architecture aside, AMD developed something else: a cohesive, merged company. Out of the pressure and pain of Fusion development emerged a different company than either of the two that had gone into it. The old days of talking about "red" and "green" teams were finally gone.

"We were similar in that we were both in a major fist fight with one guy," says Macri. "I think ATI had the fairer fight in that we were up against a similarly-sized company [Nvidia]. But this had a lot of impact on design and implementation cycles. Now, the guys at AMD had won a number of times, but it was more like David and Goliath [Intel]. It was like, 'Wow, we actually beat Goliath!' With ATI, we’d been in a fist fight for many years with Nvidia, and we won as many as we lost. So we had a different attitude about winning. ATI needed to learn that there were some Goliaths out there, and you have to be pretty damned smart to beat a Goliath. AMD learned that it actually was a Goliath in certain cases. It could be an equal. Now merge that with some faster time to market strategies. Today, our product cycle time is faster than ever across the board. So the melding gave both sides a better ability to attack not just their traditional competitors but also new competitors coming up. And those new guys coming up aren’t big. They’re all kind of small. They’re all AMD-sized. I don’t think AMD ever would have had the right attitude on how to beat someone their own size without inheriting ATI. And I don’t think ATI could have figured out how to beat someone several times bigger without AMD’s attitude of asking how you aim where the other guy’s not aiming."

Just as the two organizations were completing their cultural fusion, the Fusion effort itself was nearing the end of its first stage. AMD showed its first Fusion APUs to the world at CES in early 2011, and product started shipping shortly thereafter. In the consumer space, the Llano platforms, based on the 32 nm K10 core, arrived in the A4, A6, A8, and E2 APU series. Another announcement from 2011 CES was that the Fusion System Architecture would henceforward be known as the Heterogeneous System Architecture (HSA). According to AMD, the company wanted to turn HSA into an open industry standard, and a name that didn’t reflect a long-standing AMD-centric effort would help illustrate that fact. This would prove to be the first hint of AMD’s even larger aspirations.

Heterogeneous Roots

In the end, did Fusion matter? Quite simply, it changed the direction of modern mainstream computing. All parties agree that discrete graphics will remain firmly entrenched at the high-end. But according to IDC, by the end of 2011, nearly three out of every four PC processors sold were integrated, hybrid processors—APUs, as AMD calls them. AMD adds that half of all processors sold across all computing device segments, including smartphones, are now what it refers to as APUs.

Ubiquitous as that might sound, though, the APU is not the endgame; it’s only the beginning. Simply having two different cores on the same die may improve latency, but the aim of Fusion was always to leverage heterogeneous computing in the most effective ways possible. Having discrete CPUs and GPUs each chew on the application tasks best suited to them is still heterogeneous computing. Having those two cores present on the same die is merely an expression of heterogeneous computing more suited to certain system goals, such as optimizing high performance in a lower power envelope. Of course, this assumes that programs are being written to leverage a heterogeneous compute model—and most are not.

Ageia was one of the first companies in the PC world to address this problem. In 2004, a fledgling semiconductor company named Ageia purchased a physics middleware company called NovodeX, and thus was born the short-lived field of physics processing units (PPUs), available on third-party standalone cards. For games coded to leverage Ageia’s PhysX engine, these cards could radically improve physics simulation and particle motion. PhysX caught on with many developers, and Nvidia bought Ageia in 2008. Over time, Nvidia phased out the PPU side of the technology and supported PhysX on any CUDA-ready 8-series or newer GeForce card.

Ageia’s fame drew the attention of Dave Orton and others at ATI. Even before the AMD merger, ATI had been working to enable general-purpose GPU computing (GPGPU) in its Radeon line. In 2006, the R580 GPU became the first ATI product to support GPGPU, which the company soon branded as Stream. The confusing nomenclature of Stream, FireStream, stream processors, and so on gives some indication to the initial lack of cohesion in this effort. Stanford’s folding@home distributed computing project became ATI’s first showcase for just how mind-blowing the GPGPU performance advantage could be.

The trouble was that Stream never caught on. Nvidia seized its 2006/2007 execution upswing, capitalized on the confusion reigning at AMD at that time, and solidified CUDA as the go-to technology for GPGPU computing. But this is a bit like describing a goldfish as the hugest creature in the tank when all of the other fish are guppies. Despite a lot of notoriety in gaming and academic circles, GPGPU development remained very niche and far from mainstream awareness.

"AMD has been promoting GPU compute for a really long time," says Manju Hegde, former CEO of Ageia and now corporate vice president of heterogeneous applications and developer solutions at AMD. "But eight years ago, it wasn’t right. Five years ago, it wasn’t right. Now, with the explosion of the low-power market, smartphones and tablets, it’s right. And for developers to create the kinds of experiences that normal PC users expect, they have to go to GPU computing—but it has to be based on something easy like HSA."

OpenCL And HSA

The slow take-off of GPGPU computing had less to do with niche markets than it did with problematic programming. Simply put, the world was built to code compute for CPUs, and shifting some of that code over to GPUs was anything but straightforward.

"Various specialized hardware designs, such as Cell, GPGPUs, and MIC, have gained traction as alternative hardware designs capable of delivering higher flop rates than conventional designs," notes IEEE author D.M. Kunzman in the abstract for the paper Programming Heterogeneous Systems. "However, a drawback of these accelerators is that they simultaneously increase programmer burden in terms of code complexity and decrease portability by requiring hardware specific code to be interleaved throughout application code...Further, balancing the application workload across the cores becomes problematic, especially if a given computation must be split across a mixture of core types with variable performance characteristics."

Not surprisingly, all of the traditional APIs built to interface with GPUs were designed for graphics. To make a GPU compute math, one had to pretend operations were based on textures and rectangles. The great advance of OpenCL was that it dispensed with this work-around and provided a straight compute interface for the GPU. OpenCL is managed by the non-profit Khronos Group, and it is now supported by a wide range of industry players involved with heterogeneous computing, including AMD, ARM, Intel, and Nvidia.

So, if OpenCL provides a software framework for heterogeneous computing, that still doesn’t address the hardware side of the problem. Whether discussing servers, PCs, or smartphones, how should the hardware platform (distinct from the CPU, GPU, and/or APU) perform heterogeneous computing? Clearly, platforms were not designed for this paradigm in the past. The computing device typically had one system memory pool, and the programmer has to copy data from the CPU memory space to the GPU memory space—within the same pool—before the application can start executing its process. That same is true for fetching the results back again. In a system with only one memory pool, repetitive copying of data to different areas within the same memory is highly inefficient.

This is where HSA comes in. HSA brings the GPU into a shared memory environment with the CPU as a co-processor. The application gives work directly to the GPU just as it does to the CPU. The two cores can work together on the same data sets. With a shared memory space, the processors use the same points and addresses, making it much more efficient to offload smaller amounts of work because all of that old copying overhead is gone.

In addition to unified memory, AMD notes that HSA establishes cache coherency between the CPU and GPU, eliminating the need to do a DMA flush every time the programmer wants to move data between the CPU and GPU. The GPU is also now allowed to reference pageable memory, so the entire virtual memory space is available. Not least of all, HSA adds context switching, enabling quality of service. With these features in hardware, an HSA platform becomes very similar in programming style to that of a CPU.

"Shared memory makes the whole system much easier to program," adds AMD fellow Phil Rogers. "One of the barriers to using GPU compute today is a lot of programmers tell us they find it too hard. They have to learn a new API. They have to manage these different address spaces. They’re not sure when the right time is to copy the data. When you eliminate barriers like this across the board and enable high-level languages, you make it so much easier to program that suddenly you get tens of thousands of programmers working on your platform instead of dozens or hundreds. That’s a really big deal."

Focus On The Programmer

"Every programmer has their own favorite language," says Phil Rogers. "They can almost be religious about it. You don’t want to tell a programmer that he has to change the way he currently develops applications in order to deliver better experiences. HSA enables heterogeneous computing for all high-level languages over time."

Compatibility with C and C++ might have been sufficient for some, but AMD wanted to make sure everyone was covered, so it expanded HSA to work with C#, Java, and even functional languages. And because there’s only so much AMD can do on its own, the company decided to turn HSA into an open standard governed by the HSA Foundation, which boasts founding members including ARM, MediaTek, and Texas Instruments. Officially launched in June 2012, the HSA Foundation’s goal is to promote HSA-enabled platforms and software at all levels. This includes making SDKs, libraries, training, and other resources available to all programmers, often for free. From a developer perspective, the whole idea of HSA is that programmers can now easily take advantage of the heterogeneous compute model in their apps without being bound to design or write in any certain way.

"Programmers don’t just program to the metal," says AMD’s Manju Hegde. "They need proper compilers, debuggers, profilers, optimization tools, libraries. These are the tasks ahead of us, which is why we established the HSA Foundation, to drive the standard forward. A lot of the tools we do will be open sourced. This will give partners quicker time to market and lower the financial burden. When the software ecosystem sees that this is a genuine industry effort, the value of that will not be lost upon them. Because this is one of the first times that hardware companies are making changes to chip architecture to accommodate ease of software development. Tons of companies make changes because they want new capabilities, new features. But we’ve made changes simply to make it easier for programmers. That’s what’s needed to make things pervasive."

It’s important to keep in mind that OpenCL and HSA are two very different things, and it’s likely that the former will evolve to better fit the latter with time. Even without HSA, OpenCL offers a much different programming experience than it did even two years ago. For example, OpenCL 1.2 drastically reduces the amount of initialization code and other overhead code that used to be required for OpenCL. With HSA, that trend toward simplicity and performance will continue as programmers no longer need to manage two different memory spaces.

"Say a programmer uses Visual Studio today, writing C++ apps in Windows," says Phil Rogers. "There are hundreds of thousands of programmers doing that. For them, they can use this new technology in Visual Studio Microsoft released called C++ AMP, short for Accelerated Massive Parallelism. In C++ AMP, they only added two keywords to the language, restrict and array_view, and just by adding those two functions, those programs are marked for offload into GPU. The tiny change to the program gives numerous benefits when they have chunks of very parallel code in existing applications. It’s a much easier transition than one might expect."

HSA's Big Picture

With all of the focus on APUs and GPGPU, it’s easy to forget that there’s more to life than parallel code. The CPU remains a critical part of heterogeneous computing. Much of the code in modern applications remains serial and scalar in nature and will only run well on strong CPU cores. But even for the CPU, there are different types of workloads. Some loads do best on a few fast cores, while others excel on a larger number of lower-power cores. In both cases, and as mentioned earlier, applications need to be tailored to fit a power envelope for a particular device, whether it’s an all-in-one, notebook, tablet, or phone. As APUs gradually take over most (but not all) of the CPU arena, we’re seeing APUs diversify and segment in order to address these different power requirements. The difference now is that APUs seem likely to soon offer nearly twice the diversity of recent CPU families since they must address both scalar- as well as vector-based needs across device markets.

I attended AMD’s 2012 Fusion Developer Summit (AFDS) in Seattle, and, to my ears, it sounded like the last thing on anyone’s mind was the desktop market. There was a lot of buzz about AMD leveraging HSA to find better roads into the mobile markets. The biggest news was far and away ARM, the leading name in ultramobile processors, joining the HSA Foundation (remember, HSA is agnostic to architecture). This carries significant ramifications in many directions.

Everybody knows that mobile is hot and desktop is flat, at least in industry sales terms. To me, much of the messaging at AFDS seemed to reflect this, perhaps because desktop is the segment that seems to care least about the power and efficiency benefits HSA promises. So when I was able to sit down with Phil Rogers, I asked him if one of the outcomes of HSA would be a gradually increasing shift by AMD toward battery-powered devices and a leaning away from desktops.

"This is a common misconception," he said. "Power matters a lot on the whole range of platforms. On the battery situation, everybody gets it. Even with desktops, and more and more, what we’re calling desktops are becoming all-in-ones, people want to know not just how fast it runs but how quiet it is and how attractive it is as a product. We’re seeing that even gamers don’t want a box next to their leg pumping hot air on their shins while they’re playing. What they really want is a 30” screen on the wall with a PC built into it that runs fantastic. And in that environment, you do care about power. Even if you don’t care about the electricity bill, you don’t want fans whining and screaming or heating up and taking away clock speed."

At the other end of the market, servers stand to benefit greatly from HSA. Consider data centers and the continuing growth of cloud computing. With even smallish data centers now hosting more than 10 000 servers each, power efficiency continues to grow in urgency. Generally speaking, hardware costs comprise only one-third of a server's total cost over its service life. Another third is spent on electricity used, and the remaining third goes toward cooling costs. If HSA can help improve compute efficiency, allowing systems to complete tasks more quickly so they can turn off large logic blocks or entire cores, then power consumption can decline drastically.

"You can only pack processors so densely," says Hegde. "HSA allows you to process more densely and at a lower power envelope. I don’t need to tell you that CUDA has been doing HPC applications for five years. All those applications are so much easier with HSA because HSA is heterogeneous between GPU and CPU. Nvidia always leans toward the GPU. And certainly, there are some embarrassingly parallel applications out there, but the vast majority of applications are not, including many HPC applications. So don’t think that HSA is just about client; it’s an architecture that spans many platforms."

More About The Big Picture

For many PC users, "integrated" chips are synonymous with lower performance. This perception perhaps remains from the days of northbridge-based graphics, which were mediocre even on the best days. Most likely, that stigma will soon vanish. We don’t have the side-by-side data to show how CPU Die X performs in both APU and graphics-free versions. But it seems safe to say that in a heterogeneous environment, leveraging software designed for APU-type processing, an APU will deliver better total performance than if those CPU and GPU cores are separated into discrete components. If the APU paradigm wasn’t inherently better, the entire industry wouldn’t be shifting to it so rapidly.

Testing at Tom’s Hardware shows that, in a toe-to-toe battle between AMD and Intel, just looking at x86 performance, Intel is today’s clear victor. Far less clear is what happens once heterogeneous graphics are factored in. Just as Intel has the stronger CPU, AMD obviously has the stronger graphics architecture. What happens when the software running across these platforms takes fair and ample use of OpenCL and other heterogeneous architectures? That’s what our ongoing heterogeneous compute series is looking to answer.

"Today," says AMD’s Hegde, "the CPUs in our APUs may be lower in performance in your tests against Intel. But that’s not an indication of where we are going. We’re now embracing the low-power space in a very strong way, so we are building CPUs that are going to have very good performance with a very low power envelope. So when we say ‘balanced platform,’ we’re not saying that you have to take one or the other. We’re saying that, in a balanced platform, for every workload, do it in the place that makes the most sense. Intel’s approach of doing everything on the CPU is just wrong, because when you put a balanced workload on just the CPU, your power draw is dramatic and unnecessary. That’s why we don’t think that’s a good model. We’re going to make the cost of transition from CPU to GPU next to nothing. Then it’s up to each application to choose the right execution engine in terms of performance and power."

Similarly, AMD feels that HSA may be its path to success in smartphones and tablets, because when applications are properly optimized for balance, the GPU becomes much more influential in determining battery life. HSA targets two things: ease of programming and performance per watt. Once GPGPU compute comes into play, analysis shows that the GPU is 4x to 6x more performance per watt efficient than a CPU. ARM may have the biggest piece of that heterogeneous smartphone opportunity, but AMD is betting there’s still plenty of room for a strong number-two player who just happens to be using the same programming architecture as ARM. As mentioned, the implications for Nvidia and ultramobile would-be contender Intel are significant.

HSA Tomorrow

I asked Manju Hedge what part of the whole HSA effort keeps him awake at night. He answered without hesitation:

"How we’re going to get adoption. Our plan is to go broad. With the HSA Foundation, we made a huge bet. We’re contributing...I don’t know how many millions of dollars of IP. The reason we’re donating this to the industry is we want a rising tide to lift all. When that happens, the software ecosystem gets excited and that will be a catalyst to increasing adoption."

AMD is dead serious about this. Today, all of the company’s OpenCL tools are freely distributed. To my knowledge, every other company in the heterogeneous space charges for its tools. AMD wants all kinds of developers, from the largest names to one-person garage outfits, to be free from worrying about the economics of tools for developer environments.

Will HSA complete the reversal of AMD’s misfortunes and restore the company to its former glory? A cynic might answer: it sure couldn’t hurt. But the more thoughtful answer is that HSA is the culmination of a heterogeneous strategy born almost a decade ago. Dave Orton and Dirk Meyer saw the inevitability of today and set out to make that future happen in a way that was as beneficial to everyone involved. We see time and again that the industry gravitates to open standards and increasing efficiency. Given this, it seems likely that AMD has finally scored the victory it’s sought for so long.

"At the end of the day, we’re not maniacally focused on beating a single company," says AMD’s Joe Macri. "We’re maniacally focused on up-leveling the experience of all consumers. By doing that, many of our competition will fall out. They’ll not have the IP, or they’ll have only some of the IP. Or maybe they’ll be forced to merge, which is a very difficult thing. Intel is a wildcard. No doubt about it, they’re a very capable company and a great bunch of folks. But when you’re the incumbent, it’s much more difficult to embrace change. You only want to embrace unavoidable change because everything else costs money. So we gotta see how they pull their triggers. But I’d hate to be in their position. You don’t know how many people I interview from Intel specifically because their research doesn’t get utilized, because a VP says, 'Hey, that’s just going to cost us money. Maybe we’ll utilize your research somewhere over the next ten years.' That kills an engineer.”

At a time when cracks are appearing in Moore’s Law and the costs of shrinking fab processes continue to skyrocket, an industry that wants to keep accelerating compute capabilities must turn increasingly to optimizing efficiency. Ultimately, this is what GPGPU and HSA enable. By old methods, how much would a CPU need to evolve in order to facilitate a 5x performance gain? Now, such gains are possible simply through hardware and software vendors adopting an end-to-end platform such as HSA. No pushing the envelope of lithography physics. No new multi-billion-dollar factories. Just more efficient utilization of the technologies already on the table. And through that, the world of computing can take a quantum leap forward.

With the release of AMD's Catalyst 12.6 Beta drivers on Thursday, the company also announced that it's moving away from its monthly Catalyst release plan.

ATI originally released the first Catalyst driver in June 2002 just before its Radeon 9700 "R300" graphics card hit the market. But months after the launch, ATI decided to release a new driver each month. This continued until now, with AMD now having decided to release drivers when it makes sense.

"Our goal is to ensure that every Catalyst release delivers a substantial benefit to our end users; as we have today with the release of the Catalyst 12.6a Beta," AMD said on Thursday. "We will still continue with the Catalyst naming convention; Catalyst: Year.Month., You just won’t see a new driver every single month. We are confident that this will only benefit the end user; you’ll only need to upgrade to a new Catalyst driver, when it makes sense. "

In addition to the announcement, AMD also launched a new AMD Issue Reporting Form. The company said it made a number of improvements to the older Catalyst Crew Feedback form, making it more comprehensive, ensuring that AMD gets the best feedback possible.

"Feedback is very important in every product company’s life, especially in our fast paced environment," AMD said. " This is not just a check-box for us – we take these reports very seriously. We will review every report posted here and investigate every issue encountered. Please use this form whenever you encounter something worth mentioning."

To download the new AMD Catalyst 12.6 Beta driver, head here -- the page lists resolved issues and feature highlights. AMD customers can also download the AMD Catalyst feature preview driver for the just-released Windows 8 Release Preview.

"Featuring full support for the new WDDM 1.2 driver model, this new driver enables customers to take full advantage of the model’s features as applications are updated to leverage them," reads the driver description.

Features include support for the AMD Radeon HD 7900 Series, AMD Radeon HD 7800 Series, and the AMD Radeon HD 7700 Series, native stereo 3D support, a unified video API and more.

Just weeks after AMD's CTO of graphics Eric Demers left the company to pursue other opportunities, the company's Director of the Client Technology Unit, Godfrey Cheng, has also jumped ship after serving 12 years.

Cheng was recently part of AMD's high-profile Llano launch. But long before that, he was one of the principle players behind ATI's All-in-Wonder series of graphics cards and TV tuners which debut in 1996. He headed the project from 1998 to 2006, and then moved over to gaming technologies like Avivo and CrossFire for a year as ATI's Director of Marketing, Platform Technologies.

In 2006 he also became AMD's Director of Technical Marketing, Graphics Products where he was involved in roadmap and technical marketing for 3D graphics, video, display, power and audio technologies. He was also a technology and product evangelist responsible for benchmarking and hardware certification.

As AMD's Director of Client Technology Unit, Client Division in 2010, he managed and drove technologies such as x86, parallel compute, video and graphics for AMD's PC Client group. He also managed the competitive analysis team, the AMD performance lab, and contributed to AMD's M&A strategies.

"Marketing and Product Executive with extensive experience with PC products with focus in the areas of CPU, graphics, and video technologies," reads his LinkedIn summary. "Key strengths include product management, technology management, product marketing, extensive business development and negotiation experience working with all key Original Equipment Manufacturers, Distributors, Retailers and ecosystem partners. Passion for developing and productizing technologies that literally change the world."

The reasons for his departure is officially unknown, but he follows another former ATI employee (Demers). Rumor has it that he parted ways with AMD because it became clear the company wasn't going to follow Rick Bergman's ideas for how to best leverage AMD's graphics and APU assets. AMD has reportedly acknowledged the departure, and understands the possible concern for his timing after the launch of the GTX 680 and Demer's departure earlier this year.

"The company affirmed that it remains entirely committed to its GPU technology at all market segments," reports HotHardware. "AMD believes its New Zealand and 'Sea Islands' architectures will be a potent challenge for Nvidia's GK104 and noted that the Radeon HD 7000 family competes very well against Kepler's reduced GPGPU performance."

So far AMD has not released an official announcement.

AMD has released the updated Catalyst graphics drivers suite, now up to version 12.3. Find the download for your system here.

Highlights of the AMD Catalyst 12.3 Windows release includes: 

NEW FEATURES

Full Support for AMD Radeon HD 7000 series of products 

• Catalyst 12.3 now fully supports the AMD Radeon HD 7900, AMD Radeon HD 7800 and AMD Radeon HD 7700 series of products 

RESOLVED ISSUES

Resolved Issues for the Windows 7 Operating System

This section provides information on resolved known issues in this release of the AMD Catalyst 12.3 software suite for Windows 7. These include:

• AMD Catalyst Control Center startup issues/ delays have been resolved
• Tom Clancy HAWX: No longer crashes after task switching with MLAA (Morphological Anti-Aliasing) and Anti-Aliasing enabled
• Furmark Benchmark: No longer crashes when launched in High Performance mode.
• Quake 4: No longer display corruption after task switching when run in High Performance mode
• XPlane: Textures no longer exhibit flicker and corruption
• Skyrim: No longer displays flickering and texture corruption.
• Alan Wake: A system crash is no longer experienced when running the game in DirectX 9 mode with Crossfire enabled
• Far Cry 2: A random system hang is no longer experienced when running the game at high in game setting

Resolved Issues for the Windows Vista  Operating System

This section provides information on resolved known issues in this release of the AMD Catalyst™ 12.3 software suite for Windows Vista. These include:

• AMD Catalyst Control Center startup issues/ delays have been resolved
• Enemy Territory Quake Wars : no longer exhibits random corruption

Check out the full release notes here, which include known issues.

_{Read more from @MarcusYam on Twitter.}

Former AMD employees told Forbes that the company approached Nvidia about a possible acquisition long before it pulled out the wallet to purchase ATI for $5.4 billion back in 2006.

At the time, AMD was out to get ahead of rival Intel by grabbing a piece of the GPU market and fusing the technology with its own CPU offering. Thus AMD would be in a position to provide a CPU/GPU solution that Intel really wasn't ready to offer.

But during the talks, Nvidia Chief Executive Jen-Hsun Huang reportedly insisted that he become chief executive of the combined company. Naturally AMD Chief Executive Hector Ruiz wasn't going to let that happen, so he chose to acquire rival ATI instead.

After the acquisition, AMD struggled to integrate its newly acquired graphics business as Nvidia unleashed a flood of strong products, consuming large chunks of market share. AMD eventually gathered its forces and fought back, but Nvidia had already moved aggressively into the mobile SoC market by introducing its ARM-based Tegra chip.

With Tegra installed in tablets and smartphones, Nvidia now has a market capitalization of $9.7 billion whereas AMD is worth just $5.2 billion.

Forbes said Nvidia declined to comment.

What Does Heterogeneous Computing Really Promise?

No one is ready to declare the age of CPUs over. After all, companies like Xilinx still sell application-specific programmable logic devices that are far less functionally integrated and multi-purpose than modern central processing units. Sometimes, simpler is more effective. It’s likely that specialized processors will continue to enjoy success in certain markets segments, especially where lots of performance is the top concern. In an increasingly diverse range of mainstream environments, though, we expect that heterogeneous computing—having many types of computational resources packed onto a single, integrated device—will continue to become more popular. And as manufacturing devices, these devices will get more complex, too.

The logical endgame of heterogeneous computing is a system-on-a-chip (SoC), wherein all (or at least many) major circuit systems are integrated into one package. As an example, AMD’s Geode chips (currently powering the One Laptop Per Child project) have evolved from 1990s-era SoC designs. While many SoC products still lack the horsepower to fuel a modern, mainstream desktop PC, both AMD and Intel sell architectures that combine CPU cores, graphics resources, and memory control. These accelerated processing units (APUs), as AMD calls them, meet and even exceed the performance levels expected from typical productivity-oriented workstations. Most notably, they complement familiar processor designs with many, many ALUs typically used to accelerate 3D graphics. These programmable resources don't have to be used for gaming, though. Many other workloads are very parallel by nature, and throwing an APU with hundreds of cores at them (rather than just a CPU with two or four cores) launches an interesting debate about the potential of software optimized for highly-integrated SoCs.

Historically, on-board graphics solutions were enabled by logic in the chipset’s northbridge. Hamstrung by severe bottlenecks and latencies, at a certain point it simply became more difficult to scale up performance using platform components so far away from each other. As a result, we've seen that functionality migrate north into the CPU, creating a new breed of products able to not only offer significantly better gaming performance, but also to tackle more general-purpose tasks that leverage the hybrid nature of SoCs with CPU and GPU functionality. 

For AMD, this marks the long-sought culmination of the company’s Fusion initiative, which was presumably the driver behind AMD’s 2006 acquisition of ATI Technologies. AMD saw the potential for its CPUs and ATI's graphics technology to supplant pure CPUs in an ever-increasing share of the market, and the company was determined to be in the vanguard of that transition. Intel, of course, employs its own in-house graphics technology, but to a different end. Decidedly, its emphasis has been more focused on its processing cores and less on graphics technology.

Early 2011 witnessed the first family of AMD C- and E-series APUs arrive, manufactured on a 40 nm process. The use of integration enabled low-power 9 and 18 W models that went into ultra-portable notebooks. Today, we have the Llano-based A-series APU family. The use of 32 nm manufacturing makes it possible to cram in enough resources for a true desktop-class architecture at a value-oriented price point.

While there are a variety of specifications in play here, perhaps the biggest differentiator among the models listed below is their respective graphics engines. The A8 employs a configuration that AMD refers to as Radeon HD 6550D. It consists of 400 stream processors, Radeon cores, or shaders, whichever name you like to use. The A6 steps down to the Radeon HD 6530, boasting 320 stream processors. And the A4 scales back to a Radeon HD 6410D with 160 stream processors.

We've already run sub-$200 CPUs and APUs through a number of our favorite game benchmarks, so we know how the latest chips soar or sink in modern titles. Now we want to have a look at some of the other ways enthusiasts are able to take advantage of compute resources, though, using workloads that tax conventional CPU cores and the programmable processors found in graphics-oriented products.

In this initial installment of what will be a nine-part series, we’re putting video post-processing under the microscope. Back in the day, this would have been a time-consuming usage model, even with a multi-core CPU under the hood. Because it's a largely parallel workload, though, accelerating it with the many cores of a graphics processor has become a great way to increase productivity and improve performance.

We enlisted AMD's help in putting this series together, so we're going to focus on the company's hardware to create some pretty basic comparisons. How does a CPU on its own work in OpenCL-enabled software? How about one of the Llano-based APUs on its own? Then we'll match the cheaper APUs and pricier CPUs up to a couple of different discrete cards to chart how performance scales up and down across each configuration.

DirectCompute And OpenCL: The Not-So-Secret Sauces

On and off over the past couple of years, we've tried to evaluate AMD's efforts to advocate Stream and the implementations of Nvidia's CUDA infrastructure. Ultimately they're tough comparisons to make, though. Proprietary efforts wind up limiting the number of components you can match up against each other in a severe way. We were able to combine Stream, CUDA, and Quick Sync into one transcoding piece thanks to the diligent engineering of CyberLink and Arcsoft: Video Transcoding Examined: AMD, Intel, And Nvidia In-Depth.

Since then, AMD has moved away from a proprietary approach to general-purpose GPU (GPGPU) processing in favor of industry-standard DirectCompute and OpenCL APIs. With these, developers can more easily take advantage of the GPU’s programmable logic to perform highly parallelized tasks faster and often more efficiently than an x86 CPU on its own. Such tasks often exist within graphics-intensive workloads, but developers are gradually expanding how GPUs—and now APUs—can be applied in other areas. In fact, APUs may turn out to be a more optimized solution because they feature silicon that is ideal for both single data item (SISD) and multiple data item (SIMD) processing on the same die. Whereas applications used to stress one data type or the other, we’re now seeing increasingly graphical interfaces applied to structured data software, making a hybrid approach to processing more forward-looking. Nvidia, in comparison, is still pushing CUDA hard. But it isn't ignoring OpenCL; the company's drivers incorporate OpenCL 1.1 support.

In late 2006, ATI began catering to developers who wanted to dive more deeply into SIMD, vector-based, highly parallel computing tasks. Soon, the ATI Stream SDK and Brook+ language started providing tools that let software vendors get, as ATI said, “closer to the metal” in graphics processors. But a broader, standards-oriented approach was needed. This is where the Windows DirectX API called DirectCompute and its counterpart from the Khronos Group, OpenCL, came into play. As with DirectX and OpenGL, Windows-based apps are likely to adopt DirectCompute while OpenCL has a more platform-agnostic design.

With standard APIs on the table, developers are finally comfortable adopting GPU/APU acceleration in ways that simply didn’t happen when AMD and Nvidia were each pursuing their own competing interests. To give you a taste of what’s on deck in this article series, we're going to be exploring graphics hardware-based acceleration in:

• Video post-processing
• Gaming
• Personal smart cloud apps
• Videoconferencing
• Video editing
• Media transcoding
• Productivity and security software
• Photography and facial recognition
• Advanced user interface design

If AMD and Nvidia are to be believed, we should expect to see GPU/APU acceleration spread through a more diverse range of applications, introducing significant performance gains. Will more expensive graphics cards or more complex APUs deliver better results? Probably. Thousands of stream processors should naturally do more work in less time than hundreds. But even modest mainstream APUs should deliver quantifiable benefits.

Note that AMD’s architecture allows for the APU and certain discrete GPUs to work in tandem, much like CrossFire or SLI. So, it should be possible to start on a budget and scale up acceleration down the road. We don't really touch on this multi-GPU functionality here today, but we might as the series progresses.

Test Setup And Benchmarks

In planning this series, we asked ourselves what readers really need to know about exploiting DirectCompute/OpenCL acceleration. Is it difficult? No. The drivers enable functionality by default, and most applications able to leverage the improvements have what amounts to an on/off switch to either use the feature or not. We’re stumped as to why anyone would disable acceleration, but it does make our job of testing the features much easier.

Right out of the gate, we're testing two OpenCL-enabled post-processing applications: ArcSoft’s Total Media Theater (TMT) 5.2 (in pre-release at our time of testing) and MotionDSP’s vReveal.

The SimHD component of TMT now uses OpenCL and GPU-based processing to interpolate standard-definition video (480p) to near-HD levels (720p) in real-time. To test this, we played a DVD copy of Minority Report with (GPU) and without (CPU) OpenCL enabled. We ran in comparison split-screen mode with SD on the left half of the image and near-HD on the right half. ArcSoft provides four main features within SimHD—upscaling, dynamic lighting, denoise, and smoothness—but we only tested with the first three enabled and set to maximum. The smoothness option was not available when testing in a CPU-only processing scenario, so we omitted it to have consistency across our test parameters. Also, we only tested SimHD with AMD's Radeon HD 5870 because, as of our testing, Total Media Theater wouldn't cooperate with AMD's Radeon HD 7970. That's not surprising, considering we also had trouble getting GPU-accelerated compute working in CyberLink's Media Espresso in our AMD Radeon HD 7950 Review: Up Against GeForce GTX 580 coverage.

MotionDSP’s vReveal gained fame as one of the first and best consumer-oriented applications for fixing shaky video. The amount of processing required to pull this effect off in real-time is formidable, since several frames of video must be analyzed at once and, many features tracked and recompiled across those frames. At 1080p, this load can cripple some systems. Today, vReveal 3 also includes several additional features, including sharpening, brightening, and noise cleaning, all of which can run concurrently with stabilization. Ideally, you'd have all of this rendering in real-time.

In our tests, we used two sample video clips that ship with the downloadable binary of vReveal 3: the “Barcelona” file at 480p and the “San Francisco” file at 1080p. We tested these twice, once with only basic stabilization enabled and then in a more demanding configuration with five effects piled on. We then tested these in CPU-only, APU-accelerated, and two different discrete GPU-based configurations.

Tom's Hardware Q&A With ArcSoft

ArcSoft’s Total Media Theater is one of the top names in PC-based movie playback, and the SimHD functionality within the application, which enhances video with resolution upscaling and other features, has long been an early beneficiary of GPU-based hardware acceleration. Kam Shek, director of technical marketing for ArcSoft, sat down with Tom’s Hardware to provide an inside glimpse of his company’s process in making the switch to OpenCL-based acceleration.

Tom's Hardware: Many people fail to appreciate the expense and hard work required of hardware vendors in order to enable next-generation software. Obviously, the graphics vendors have a vested interest in popularizing OpenCL. How did that interest translate into action with ArcSoft?

Kam Shek: There are a couple of things we’ve done in optimizing our software. For AMD specifically, we added support for the fixed-function UVD 3 decode logic that includes hardware acceleration for the MPEG-4 ASP [Advanced Simple] profile. We also have our SimHD technology, which is our in-house upscaler. We recently ported that to the OpenCL language. In the past, we used ATI Stream, but about six months ago, we started the port to the new industry standard, working very closely with the company on that.

TH: Why did you choose to support OpenCL?

KS: Because OpenCL provides a heterogeneous environment for us, meaning we can actually program on OpenCL regardless of what part of the algorithm might be running on the CPU or GPU. As a result, we can have a much better balancing of the system resources when we schedule our algorithm between those two.

TH: In moving from proprietary approaches like Stream and CUDA to OpenCL, what advantages do you see in the new technology over the old?

KS: Well, one of the most important is that OpenCL is an open standard. ATI Stream was proprietary. Second, OpenCL allows us to scale better between the CPU and GPU. Let me explain a little more what we do with SimHD. SimHD is about more than upscaling. It has multiple processing units in it. One is for scaling, but then there’s a denoise unit, deblocking, sharpening, dynamic lighting, etc. So some of the algorithms by nature run better on CPU. For instance, if there’s case switching, which happens a lot in dynamic lighting, that’s more efficient on the CPU. But for floating-point, high-precision calculations, those actually run better on the GPU. So we schedule our algorithm between which function runs better on which processing core, then we balance accordingly.

TH: Do you have a sense of the optimization ratio, CPU versus GPU?

KS: I think we’re seeing, on our AMD system here, CPU usage drop by about 10% and then the GPU utilization increases about 20%.

TH: So increasing GPU usage by 20% saves 10% off your CPU utilization?

KS: I’d say 15% to 20%, yes, but I’d probably need to look up the exact number.

TH: Are there any considerations that must be made when coding for an APU compared to a discrete GPU?

KS: No, from an OpenCL perspective, it’s all the same. Now, before we run the algorithm, we do query the capability of the graphics. If the graphics have more Stream cores available, then we’ll run more algorithms on them. But in general, it’s the same. We don’t need a different binary for discrete graphics versus an APU. However, I should add something. As I said, the GPU is better for many floating-point, high-precision tasks. The CPU version for these algorithms are somewhat streamed down, so the precision is not as good as running on the GPU. The algorithm for the GPU is more complex than the one for the CPU, so it provides better precision.

Tom's Hardware Q&A With MotionDSP

In what seems forever ago, this author had his first brush with MotionDSP on Tom’s Guide while trying to find out if the photo/video analysis done in movies is actually possible. Can someone take blurry, small footage and process it into a clear, recognizable face? MotionDSP’s vReveal is the consumer version of its Ikena software, which caters to vertical markets, including government agencies. The company’s algorithms can provide some rather amazing results (although there are limits), but this sort of post-processing takes a heavy hardware toll. What should you expect when getting into this sort of processing? We talked to MotionDSP CTO Nik Bozinovic and CEO Sean Varah to find out.

TH: How did you guys get started with hardware-based acceleration?

SV: When we were starting five years ago, there wasn’t an easy way to program GPUs. Basically, as we were working on our algorithms as early as 2008, we realized that as video was moving from sub-SD to standard-def to HD, there would be an increased need for high-performance computing. Only with that can we process video in real-time, which is absolutely mission critical on our professional-grade product. So the need to use heterogeneous computing was very obvious long before it became a reality, before vendors like AMD and Nvidia started supporting it. It’s been on the road map for years, but it’s only been in the last 12 to 24 months that we’re finally seeing that promise of having supercomputer-like performance from a GPU become a reality. Honestly, it’s really making a big impact on our bottom line.

TH: What are you doing with heterogeneous computing capabilities in your software?

SV: A couple things. One is...there are several simultaneous things that are wrong with video at any one time. It’s very rare that you have the perfect camera in perfect conditions, especially with consumers. There could be problems with resolution, noise, lighting, stabilization. So, in vReveal, we’ve packaged a series of different video filters that together can address these problems, and we’ve made the process incredibly simple by putting them all into an automatic, one-click fix operation. The reason we can make that a snappy experience for consumers is because we utilize heterogeneous computing.

NB: In addition to stabilization, we have noise enhancement, or noise removal, which we call a cleaning filter. We have auto light balance, sharpening, contrast improvement, and, as Sean mentioned, all this can happen automatically, with the complexity hidden from the user. But this is where heterogeneous computing comes in extremely handy. We have advanced video processing and video analysis tools that are all taking advantage of heterogeneous computing. I’ll just give you a couple of examples. In vReveal, we can, in near-real-time, create panoramas out of panning videos. You can take any panning shot, click a button in vReveal, and just mere seconds later end up with a beautiful stitched panorama in our pro-grade software called Ikena. We have a similar thing where stitching happens on the fly, and you can create massive mosaics used for different things, and that wouldn’t be possible without using GPU. Sure, we started as a video enhancement company, but now, using these GPU capabilities, we’re way beyond that.

TH: Specifically, what aspects of your software wouldn’t be possible without GPU-based acceleration?

SV: Well, doing it in real-time wouldn't be possible without the GPU.

NB: It’s especially true with higher-resolution media because, aside from just the sheer compute bottleneck that you solve by going to GPU—or heterogeneous computing in general, as opposed to running things only on the CPU—you are also solving a bandwidth bottleneck problem. What I mean by that is this: for our software to work and create the desired results, we have to look at a number of frames at the same time—from a couple to 30, 40, or 50 frames. It’s a very memory- or bandwidth-intensive problem to even a larger degree than it is a compute-bound problem. You have a double-win when you execute something like that on a GPU because, just to copy a large number of massive, uncompressed, high-definition frames is something you can do on a GPU, but it’s impossible on the CPU. It’s almost an order of magnitude difference between the memory bandwidth on these two devices.

Tom's Hardware Q&A With MotionDSP, Continued

TH: Not that long ago, we were seeing test results that showed how users got most of their performance pop simply from having GPU-based acceleration, not necessarily from dropping a small fortune on a top-end card with GPGPU support. Has that changed? Because with APUs, that support is essentially now built-in.

SV: What’s changed is companies like AMD now are shipping these Fusion chips that combine the GPU with the CPU on the same die. That makes a difference. You’re right—before, if you had a discrete GPU, you’d get the most bang for your buck just having the GPU there. But having that discrete GPU meant you were going to be buying a higher-end laptop or desktop. What’s cool about the Fusion chips is that you’ve got a pretty darn good integrated GPU in, say, the Llano platform. For a $500 or $600 price point, a consumer can get a laptop that seriously kicks ass with our software. Before, they would have had to buy a $1000 laptop with a discrete GPU to get the same level of performance.

NB: Let’s try to provide a different metric. A high-end, modern CPU from AMD or Intel will retail in the $200 to $300 range. You can outperform that CPU by a factor of 3x to 5x with a GPU that costs about the same. If you have a discrete GPU that costs about $300 and you run our software on that component, you’ll get a three to five times higher frame rate. Now let’s say you want to spend $1000 on a CPU. As I said, there are certain bottlenecks where massive multi-frame processing cannot happen on that CPU because of memory bandwidth and compute bottlenecks. You’re hitting the limits, and it really becomes irrelevant how much money you’re spending on the CPU. You may be hitting a threshold of six or seven frames per second and you can’t go any faster. That's just because you are hitting certain inherent limits on the CPU the GPU doesn't have.

SV: Nik and I are saying two different things. I’m saying on the consumer side, on a $500 or $600 laptop with Fusion chips like Llano, you’re getting the performance that recently cost $1000 in a laptop. And on the discrete side, he’s saying that for the same price, you’re going to get like five to eight times the performance per dollar from a GPU compared to spending that money on a CPU.

NB: If you go to standard def-video, that’s where CPUs in general fare better than GPUs. But everybody takes 720p or 1080p now with their Android phones and iPhones, so the difference between CPU and the GPU becomes much, much larger. As Sean said, an off-the-shelf laptop you buy for $600 is going to give you the performance of a $2000 desktop from before the heterogeneous computing era, just because you didn't have the power of parallel computing at your fingertips.

TH: We’ve got OpenCL in play from AMD, Nvidia, and Intel now. Nvidia is also still advocating CUDA as a closer-to-the-hardware solution, and Intel's approach seems to be running OpenCL on its CPU cores. As a developer, you obviously need to support as broadly as possible, but technically, how much of a difference do these rival strategies make?

NB: From an ISV perspective, we would like to have one standard that runs everywhere. CUDA, being a vendor-specific technology, is something less appealing to us than OpenCL at this time. Now, it’s a fact that CUDA was the first one out there. Nvidia's the incumbent. But we’ve honestly been impressed by the speed of development on OpenCL-based tools and the entire chain, from like SDK to tools to runtime and how things run in a driver. We’ve seen very significant commitment, and I’m talking from an ISV perspective, where things were not so pretty even 18 or 12 months ago. It was very hard or impossible at that time for us to promise to deliver a solid product using heterogeneous computing on OpenCL. That changed completely over the last year. Right now, AMD is the vendor behind OpenCL development and implementation. We don't know what Intel is going to achieve with Ivy Bridge, but ARM announced that it's going to dip its toes into OpenCL water.

Benchmark Results: ArcSoft Total Media Theatre SimHD

As noted earlier, and in our Radeon HD 7970/7950 launch coverage, AMD hasn't yet ironed out the kinks with hardware acceleration in its newest architecture. So, it wasn't much of a surprise that SimHD wouldn't work with our 7900-series boards. Instead, we dropped in a Radeon HD 5870, which is currently compatible. Its 1600 shaders operating at 850 MHz still deliver respectable performance in today's games, so we're curious to see what this now fairly-mainstream card can do in a compute workload.

In this first test result set, we ran all of the post-processing on the CPU, with hardware acceleration disabled. Low, average, and high CPU utilization values are shown (in percent) to indicate the range of processor resources consumed by SimHD’s upscaling. Both mobile platforms are roughly comparable, though AMD's average utilization is likely lower as a result of its four physical cores.

The desktop numbers are more interesting. Yes, the overall utilization is roughly two to three times less on the desktop, which is telling since both mobile processors sport 35 W TDPs compared to the 100 and 125 W APU and FX CPU.

But also, notice how there is very little difference between the A8 APU and the higher-end FX chip. Although the FX has more transistors dedicated to general-purpose processing, a nice large shared L3 cache, and higher clock rates, it isn't able to demonstrate an advantage over the more mainstream A8. Clearly, there's some other bottleneck in play.

With GPU acceleration enabled, the APU-driven notebook starts to exhibit a performance advantage over the CPU-only model, though perhaps not to the extent we were expecting. Any advantage is notable when you're talking about a battery-powered piece of hardware, however, the more significant gains start cropping up when you have the resources of a desktop-class configuration to throw at the upscaler.

Clearly, the biggest gains are available to those who use more potent discrete graphics on a platform with a fast CPU to accelerate demanding workloads. Testing the same graphics card next to AMD's A8 doesn't yield as compelling of a gain, likely due to a processor bottleneck.

When it comes to balancing out performance for your dollar, though, the A8 leaning on its integrated Radeon HD 6550D graphics engine is the real winner here. Consider that the A8-3850 sells for $135. AMD's idea here has to be one of putting two technologies together at an affordable price, where in the past you would have needed separate entry-level CPUs and graphics card to achieve the same thing (and the Radeon HD 5870 wasn't a cheap board, either).

Benchmark Results: vReveal On The FX-8150 And Radeon HD 5870

Again, we're trying to compare the experience you get before and after applying DirectCompute/OpenGL acceleration. The trick is identifying the right environment to flag as normal. Is the 480p video of DVD still an acceptable baseline, or is 1080p the very least you'd accept? In an app like vReveal, should we expect people to only stabilize their videos, or will they take the kitchen sink approach and apply every effect? Ultimately, these four variables were what we compared using GPU-based acceleration, and then turning it off.

We’re going to examine our data in terms of CPU utilization, measuring system impact, as well as render speed. Rather than indicate frames per second (which pegs at 30 and stays there, telling us very little), vReveal spits back a percentage of real-time at which a render job is operating. This is probably a more meaningful number to the average user. For instance, if a one-minute video clip is rendering at 50%, the render job takes two minutes to complete.

Predictably, we’re going to see a lot of data showing that GPU acceleration takes a large strain off of the CPU, and 1080p video is much more demanding than 480p.

The Radeon HD 5870 benefiting from OpenCL-based acceleration handles a single image enhancement on 480p with ease, offloading almost the entire task from an FX processor. But don’t assume that this means the FX is impervious. Without GPU assistance, that one effect applied to the 1080p video via software chews up almost half of the chip's processing time. Enabling GPU acceleration brings utilization down to about 10%, validating claims of a 4x to 5x benefit from hardware-based compute.

Now, what does GPU-assist mean in terms of time and getting a job completed? With only one render effect active, even our 1080p clip is processed in real-time. Note that the 480p clip can render in real-time without hardware acceleration, but the 1080p clip cannot, and requires almost twice as much time to complete.

Let’s crank up the dial and apply six render effects. Interestingly, there isn’t much difference between the 480p and 1080p clips when you use AMD's fastest CPU to manipulate them. Simply having six effects to chew on almost fully utilizes the processor, regardless of resolution.

When we enable GPU acceleration on the Radeon HD 5870, the performance improvement is even more dramatic than our single-effect baseline. We measure an 11x difference in CPU utilization with the 480p clip and 6x for the 1080p. Even still, it's interesting that the 1080p workload is twice as demanding as 480p with six effects applied.

Although this test pegs our FX-8150 at nearly 70% utilization, you can almost render the 480p clip in real-time using software mode. The same cannot be said for the more demanding 1080p clip, which slows to 13% of real-time when AMD's FX is the only resource operating on it. That means processing takes about eight minutes for every one minute of video footage you feed through.

Enabling hardware acceleration brings us back up to 92% of real-time—a 7x gain under a worst-case load.

Benchmark Results: vReveal On The FX-8150 And Radeon HD 7970

Using the same platform, we swap out the old Radeon HD 5870 in favor of the newer Radeon HD 7970 to gauge whether or not the GCN architecture has any bearing on our benchmark results.

When working with only one render effect, an admittedly lightweight metric, there isn't much difference. Even at 1080p, the Radeon HD 7970 leaves our FX processor at 12% utilization, whereas the Radeon HD 5870 dropped the FX's workload to 10%. That's right. Our data shows a slightly higher CPU load with the newer GPU.

Because this result reverses as we apply more effects, it's conceivable that the 7970's compute resources aren't being utilized as effectively under the light load. Meanwhile, the heavier burden lets the 7970's 2048 shaders stretch a bit. More important is that we’re still seeing low single-digit utilization with accelerated 480p video and a 4x performance gain with 1080p.

Understandably, there is no visible difference in render speed when switching up to the Radeon HD 7970 here. The test only goes up to 100%, and trading GPUs should have no impact on software-only processing.

Again, in terms of CPU utilization, we’re seeing almost no benefit from the Radeon HD 7970 under our heaviest vReveal load compared to the older Radeon HD 5870, despite the 7970's architectural advantages. This is still good information, though. It tells us that we can't always expect scaling that corresponds to the GPU's potency. Of course, this is going to vary by application and, in some tests, a faster graphics processor absolutely will mean better performance.

Apart from an almost imperceptible and insignificant nudge of the needle in the 480p software test, these results show the same 100% rendering seen with the Radeon HD 5870.

Benchmark Results: vReveal On The A8-3850 With Discrete Graphics

Now we edge closer to the big question of the day: is AMD’s APU platform capable enough to tackle the same demanding tasks for which we usually recommend FX and Core i5/i7 processors? Let’s warm up by examining how the CPU component of the chip fares when backed by discrete graphics.

In this context, looking at our CPU-only software results should be the most telling. Sure enough, we see the non-accelerated 480p test shows that the FX-8150 enjoys 13% lower CPU utilization compared to the A8, while the 1080p clip lets the FX cruise around at 22% lower utilization. With an almost 70% load on the A8, we’d be reluctant to even attempt this operation while multitasking in the real world.

However, notice that utilization also leaps on the A8 even when GPU acceleration is turned on. Whereas the 1080p load on our FX-8150 and Radeon HD 5870 platform was 10%, it jumps to 38% on an A8-3850 and Radeon HD 5870. This is the most compelling evidence that, even in the world of OpenCL and hardware-based acceleration, it still takes a balanced combination of CPU and GPU to yield the best experience. Offloading a task doesn't mean it's completely alleviated. Clearly, MotionDSP still relies on a capable host processor for part of its workflow.

With only one effect active, our job still renders at full speed in all cases except 1080p in software mode. Despite the significant difference in CPU utilization, the net effect on render performance between our two CPUs is only 6%.

With six effects enabled, we again see the FX offer a slight utilization advantage over the A8 when running in software. But given that all numbers are at least 67%, we’d say the benefits of the FX here are nominal.

With GPU-assist turned on, all of our utilization percentages plummet. However, the A8 still shows twice as much utilization as the FX, predictably indicating that the tasks handled by each host processor hit AMD's A8 harder than its flagship FX.

Reinforcing the point, we see the A8 take a 25% render speed hit compared to the FX when relying on a Radeon HD 5870 for acceleration. This is interesting because, while the CPU utilization numbers seem unkind to our FX in a CPU comparison, these real-world numbers show that the FX can still yield a significant time savings in high-load scenarios.

No surprises here. The A8 APU incurs 15% to 20% higher utilization than the FX.

More important, while the FX-8150 and Radeon HD 7970 combination demonstrates a 3.5x CPU utilization benefit from enabling graphics acceleration at 1080p, the A8-3850 enjoys a less than 2x benefit. Again, we see the pitfall of putting too-powerful of a graphics card with a processor that can't always keep up. Balance is critical when you want to maximize performance, and it really takes a higher-end CPU to fully exploit what a potent graphics card offers.

The 1080p accelerated test shows our A8 running at 27% utilization with a Radeon HD 5870 plugged in. It jumps to 40% with a Radeon HD 7970 installed. And yet, when we look at render performance, the Radeon HD 7970 has a 24% advantage, managing near-real-time output. What's all of that mean?

It'd appear that the 7970 is doing more work than the 5870, in turn creating a larger workload for the A8. Although you end up seeing higher CPU utilization, the end result is an experience that's closer to ideal. That is to say, close to real-time rendering is possible here where it wasn't on the Radeon HD 5870.

Benchmark Results: vReveal On The A8-3850's Radeon HD 6550D

With discrete graphics fully benchmarked, we can switch over to the A8's integrated Radeon engine and compare.

With GPU acceleration enabled, we see a slight advantage from discrete graphics when handling 480p video.

The real shocker is that CPU utilization is almost identical when the integrated and discrete graphics subsystems work on 1080p video. The explanation is simple. A lighter load doesn't tax the Radeon HD 6550D, so you get real-time rendering from the APU. The implication is that you don't even need an add-in card to get these results, creating an opportunity for very mainstream users to see some pretty cool value.

But can the same conclusions be drawn when we test six effects rendering simultaneously?

Not quite. Because we have less graphics hardware to throw at this problem, CPU utilization turns out to be pretty similar whether you're looking at 480p or 1080p video. Turning on GPU-based acceleration clearly helps alleviate the workload, dropping our A8 to about 27% utilization at both resolutions.

However, the performance chart tells the second half of the story. Pushing CPU usage down is great at 480p, where host processing and graphics working together manage real-time rendering of six effects. But at 1080p, the two subsystems are collaboratively stuck at 29% of real-time. That's less than half of what the Radeon HD 5870 was able to do matched up to AMD's APU. For serious compute workloads, the sheer complexity of a discrete GPU is undeniably superior.

Mobile Platform Results And Wrap-Up

In looking at these notebooks, we need to answer two questions. First, are the integrated graphics engines on low-power mobile processors fast enough to accelerate a compute-oriented workload? Second, how do AMD's mobile APUs compare to its desktop implementations?

First let’s compare the Gateway machine running an A8-3500M to the desktop A8-3850. Statistically, the two are almost in a dead heat, which is pretty phenomenal when you consider that the A8-3500M is a 35 W part standing up to the 100 W A8-3850.

While 480p video gets handled in real-time on both platforms, slower core and graphics clocks catch up to the notebook. It takes a hit on the 1080p test, losing one-third of its frame rate.

When we crank up the vReveal load to six effects, the system's CPU utilization tops out in the 80-85% range with software-based rendering. The A8 APU also vacillates between 26 and 32% with GPU acceleration turned on, suggesting that this platform is again being topped out based on the way it distributes CPU- and GPU-specific work.

In this demanding application, the Gateway notebook still manages to cope with 480p video in real-time (100%). Also impressive is that, in the 1080p benchmark, the mobile APU only takes a 5% hit compared to the desktop A8-3850. Although 24% of real-time seems pretty grim next to the Radeon HD 5870's ability to achieve 67% of real-time performance in this test, the fact that we're comparing a 35 W part to a 100 W APU and 188 W graphics card working cooperatively speaks volumes, too.

Finally, let's check to see how the APU fares against Intel's Sandy Bridge-based architecture with HD Graphics 3000. Of course, all of the tests are run in software mode, so we're only gauging performance without OpenCL-based acceleration.

With 1080p unaccelerated video processing, the APU-powered machine was pushed to 69% utilization with one effect and 85% with six. Here, we see 70% and 84%, respectively.

Nevertheless, Intel's mobile Core i5 is able to post decent numbers. With one effect applied to the 480p sample, it manages real-time playback. With six, it falls to 75% of real-time. Unassisted, AMD's APU was only able to hit 40% of real-time in the same test. The thing is, collaborating with its 400 Radeon cores, the A8 jumps back up to real-time rendering.

Those results are exacerbated at 1080p. In a CPU-to-CPU comparison, Intel doubles AMD's performance. But once you get the GPU acceleration involved, the APU assumes a definite lead.

Learning Lessons

After all of that, what do we come away with?

1. When time is money, there is no substitute for high-end hardware, and discrete components remain important upgrades for enthusiasts. None of the configurations we tested could touch our FX-8150 and Radeon HD 7970 working together. Had this comparison involved even pricier processors, we probably could have predicted the outcome of the software-only tests. In any case, balance remains a critical component in picking the hardware for your next upgrade.
2. When budget is a concern, getting general-purpose and graphics resources on the same piece of silicon can facilitate notable performance boosts in applications optimized to capitalize. Yes, a mid-range discrete graphics adapter will outperform as a result of its more extensive architecture. However, the APU design gets you in on the ground floor of GPU acceleration without requiring an add-on graphics card (with its corresponding expense). APUs most certainly won't satiate enthusiasts accustomed to powerful discrete adapters, but they do put a much-needed emphasis on capable integrated functionality where previous chipset- and processor-based engines came up short.
3. The benefits of a mobile APU can be close to equivalent to a desktop APU, again, in the right application. There's a reason the desktop A8-3850 has a 100 W TDP and the mobile A8-3500M is constrained to 35 W. However, there's also something to be said for an intelligent balance between processing power and graphics muscle to help notebooks yield a better overall experience.

In all honesty, this is the most exciting development we’ve seen from AMD in quite a while, and it deserves more attention. Stay tuned for our next installment on gaming acceleration in a few weeks, and we’ll see if similar (or better!) results can be enjoyed there, as well.

AMD quietly released a new set of graphics drivers last week just at the tip of the weekend. If you missed them, consider this a reminder that the Catalyst driver suite now has a dozen versions in its history.

Those awaiting stable drivers can now download the Catalyst 12.1 drivers that feature application profiles and video UI improvements. These new features first rolled out in the form of a preview driver released mid-December.

If you insist on living on the edge, AMD also released an even more advanced 12.2 preview driver set. The main improvements come to those using Eyefinity:

• Additional resolution support:   Users can now choose from a larger set of resolutions when running AMD Eyefinity
• Dynamic Configuration Changes:  Switching between different display configurations will occur automatically when physically plugging/un-plugging displays
• HydraVision™ enhancements:  The Windows® Task bar can now be moved and resized based on users preference
• Profile Manager improvements:  Increased support for Display Groups (including 5x1) and Extended configurations within the Profile Manager

Gamers will also appreciate the new version as it fixes a number of bugs when running on Windows 7.

• Call of Duty IV no longer displays random corruption when in DirectX® 9 more and the Edge Detect filter is enabled.
• A black screen is no longer occasionally displayed when running Portal 2 in 3D mode.
• All 3d Anti Aliasing settings are now correctly listed in the Vision Control Center.
• Morphological filtering settings are now correctly listed in the Vision Control Center.
• Civilization V no longer experiences random crashes when run in DirectX® 10 and DirectX® 11 modes.
• Enabling the ‘Edge Detect’ filter in Resident Evil 5 or Just Cause 2 no longer generates a random application crash.
• Battlefield 3 no longer hangs when MSAA is enabled.
• Saints Row III no longer experiences random hangs.
• Rage no longer experiences intermittent hangs when Crossfire is enabled.
• Dragon Age 2 textures no longer flicker when run in DirectX® 9 mode in Crossfire configurations with vsync enabled.
• Crysis Warhead and Battlefield Bad Company 2 no longer experience random crashes when Crossfire is enabled.
• Texture flickering is no longer experienced when playing Battlefield Bad Company 2 in DirectX® 11 mode.

You can download the Catalyst 12.2 preview drivers here.

_{Read more from @MarcusYam on Twitter.}

Mobile Model Shenanigans: What's In A Name?

Power consumption is the big neutralizer when it comes to separating desktop and mobile component design. As you know, power is dissipated as heat, and heat is public enemy number one. We certainly can’t walk around with three-inch-thick notebooks that consist of more copper than component space and require desktop-sized power supplies. Even if such a device were portable, we surely wouldn't want it resting on our...laps.

So, we're not unreasonable here. The idea that certain pieces of hardware have to be downclocked before they're deemed suitable in notebooks is acceptable to us.

But then came the naming games. The earliest any of us can remember a mobile part being upgraded (in spirit, at least) to an inflated model name was when ATI’s Radeon 9600 XT magically turned into the much larger Mobility 9700. ATI could have excused that move with the explanation that the Radeon 9600 XT outperformed an underclocked Radeon 9700 desktop GPU at similar power levels, but that was just the beginning of a long journey down the rabbit hole.

Stranger the naming became, until we finally ended up with underclocked upper-mainstream gaming GPUs getting the names of their extreme counterparts in the Radeon HD 6970M and GeForce GTX 580M. But all of that paled in comparison to one of AMD’s most recent mobile releases, the single-GPU Radeon HD 6990M.

You might expect the 6990M to consist of two Radeon HD 6970M GPUs, since that's how we get a desktop-class Radeon HD 6990.

Instead, we see AMD switch from an underclocked version of its Radeon HD 6850 to an underclocked version of its Radeon HD 6870. Whereas the desktop Radeon HD 6990 has almost two times the raw computational power of the Radeon HD 6970, the 6990M gains less than 25% over AMD's 6970M.

Both AMD and Nvidia excuse this naming disparity by affixing their flagship desktop model name to their flagship mobile products. However, we see this as a great way to deceive uninitiated notebook buyers. After all, the performance profile of each desktop GPU is trumpeted to the masses before the notebook GPUs with similar names are introduced. So, “common knowledge” works against the partially-informed buyer.

Man, just imagine if Intel called its mobile flagship the Core i7-3960XM. Though, at $1050, the four-core, 2.7 GHz Core i7-2960XM is only one digit away from following along...

Amidst all of this, some say that the other limitations of mobile computing make additional GPU power unnecessary. Notebook screens are relatively small, with moderate resolutions that typically don’t require as much graphics horsepower to drive. Although they exist, we haven't yet tested any mobile gaming platforms capable of spanning across three screens via Eyefinity or Surround, and single-screen resolutions as high as 1920x1080 aren't particularly taxing on today's graphics processors.

Nevertheless, today we’ll we compare the real performance differences between desktop and notebook GPUs while testing within the limits of a real high-end desktop replacement notebook.

Test Settings And Benchmarks

Although a desktop builder could put two dual-GPU Radeon HD 6990 graphics cards into a single system, notebook builders are stuck with two GPUs at most. MXM modules are simply too small to accommodate two large GPUs, let alone the power circuitry and doubled-up memory resources. So, we're going to put AMD’s mobile graphics in the best possible light by comparing two of its mobile Radeon HD 6990M modules to a single desktop Radeon HD 6990.

Nvidia isn’t getting a pass for its own indiscretions. We're also comparing a pair of its GeForce GTX 580M graphics cards to a single GeForce GTX 580. Nvidia's solution can't be included in as many of the tests, since we no longer have access to the mobile modules we'd need in order to run some of our newer benchmarks. That also explains why the older 280.26 display drivers were used. But it's worth noting that Nvidia is guilty of the same naming practices.

Fortunately, our legacy gaming suite include GPU-busters like Just Cause 2 and S.T.A.L.K.E.R.: Call of Pripyat.

Notebook motherboards are generally pretty limited with regard to their features. So, a budget desktop motherboard using the same chipset makes for the most accurate comparison. ASRock’s X58 Extreme3 fills the need for desktop graphics connectivity.

Benchmark Results: Crysis

Retired immediately after our previous GeForce GTX 580M review, Crysis has shown us bottlenecks in everything from GPUs to storage subsystems. GPU performance scales a little faster than CPU performance, which is one of the reasons we began looking for newer benchmarks to stress our graphics configurations.

This could be the perfect example of why so many owners of AMD cards insist they be tested on overclocked Intel platforms: the Radeons appear more CPU-bottlenecked than the GeForce solutions, and even the stock Core i7-990X CPU used by all configurations isn’t enough to overcome that shortcoming.

Big spenders don’t want to play their games at 1680x1050. But, if they had to, they’d find both of the single-GPU mobile configurations to be unacceptably slow using High quality details and 4x AA. The dual-GPU setups remain CPU-bottlenecked, for the most part.

Our highest 1920x1080 settings are where AMD's Radeon HD 6990M CrossFire finally catches up to the GeForce GTX 580M in SLI. The desktop card needs even higher resolutions to prove its superiority, but this is the point where we run out of scalability on the notebook's panel.

Both AMD and Nvidia demonstrate a single desktop card outpacing two of their notebook-oriented modules named too similarly for our comfort, which starts to illustrate our point about customers potentially being led astray as they spec-out their notebooks. A higher available resolution would have simply driven this point home more definitively.

Benchmark Results: Just Cause 2

Just Cause 2 applies a lighter CPU load than Crysis, giving every configuration a chance to achieve higher frame rates based on unfettered GPU muscle.

Desktop cards begin to show their might at 1920x1080. However, this application still leaves us wanting a higher test resolution to further illustrate the differences between mobile and desktop graphics models. Note that a single Radeon HD 6990 is more than twice as powerful as a single Radeon HD 6990M, even at the medium resolutions supported by notebook displays.

Benchmark Results: S.T.A.L.K.E.R.: Call Of Pripyat

The Call of Pripyat benchmark shows the desktop 6990’s superiority over its notebook namesake, even at a low 1280x720 resolution.

The GeForce GTX 580M SLI setup's advantage over one desktop GeForce GTX 580 sibling might appear a little strange, if not for the knowledge that the GF114-based GeForce GTX 560 Ti (on which the mobile part is actually based) does really well in SLI.

As expected, the desktop Radeon’s performance advantage grows as resolution increases. SLI continues to scale well in this benchmark. So, we're left looking at the AMD cards, wishing for even higher resolutions to push up to.

Benchmark Results: DiRT 3

We added DiRT 3 testing to our notebooks only after returning the mobile platform with GeForce GTX 580Ms installed, leaving AMD to stand on its own in this desktop-versus-notebook graphics comparison.

Ouch. We would have expected 1280x720 to be CPU-bottlenecked, but the single desktop card is already outpacing a pair of notebook modules bearing the same model number (minus that very important M, of course) by a wide margin. Buyer beware!

We think its both sad and wrong that two Radeon HD 6990Ms can’t beat one Radeon HD 6990 when complemented by the same CPU, memory speed, and chipset. Pathetic.

Benchmark Results: Metro 2033

AMD’s Radeon HD 6990M CrossFire solution is again decimated by one Radeon HD 6990 in Metro 2033.

There’s not much left to say as resolutions increase. But we can summarize by saying that the 6990M isn’t powerful enough for our highest Metro 2033 test, even in CrossFire, and even at a modest 1920x1080 resolution. Although partially-informed buyers can only blame themselves for insufficient research, in their shoes, we'd still be demanding a refund.

Benchmark Results: StarCraft II

What's wrong with our StarCraft II results? Both the dual-GPU desktop and mobile CrossFire solutions are CPU-bottlenecked, and the game appears to have a slight tendency toward the Clevo-built platform.

Cutting Through Mobile Graphics Naming

With Nvidia’s mobile solution no longer available to us when we started adding newer games to our notebook suite, we were forced to break today’s benchmarks into two groups. Performance differences in the older collection of tests were tightened by CPU bottlenecks in Crysis, while StarCraft II caused the same phenomenon in the newer group.

With a mix of resolutions starting at a unrealistically-low 1280x720, the desktop Radeon HD 6990 outpaces a pair of Radeon HD 6990Ms by around 10%. However, most readers will really want to see is how these solutions perform at a high-end notebook’s 1920x1080 native resolution.

A few folks tried telling us that CPU bottlenecks would make 1920x1080 results meaningless for pair of Radeon HD 6990M graphics modules. And still, the dual-GPU desktop card still outpaces the notebook solution by around 13% on average. And that’s with a known CPU-bottlenecked game thrown into the mix for both gaming suites.

Nvidia is only slightly less-guilty than AMD since it doesn’t give any of its mobile modules a model number equal to one of its dual-GPU desktop cards. In other words, its fastest mobile product wasn’t forced to face-off against a GeForce GTX 590. Though we don't witness the same bloodbath suffered by the AMD mobile-to-desktop comparison, it still takes two GeForce GTX 580Ms to approach the performance of a single GeForce GTX 580.

AMD rates its Radeon HD 6990 at nearly four times the power consumption of its mobile counterpart, while Nvidia’s GeForce GTX 580 desktop card draws 2.5 times the wattage of its mobile sibling. We've already heard the arguments from both AMD and Nvidia in justifying the deception: "these are our fastest mobile products, and we can't deflate the names so long as the other side is doing it too." Our assertion (and cautionary benchmark results) stand, though. Today's naming is marketing-driven, spurred by AMD and Nvidia trying to one-up each other in promoting mobile architectures, and it only continues to worsen over time. 

While the performance of these mobile solutions with dual GPUs often appears wonderful at 1920x1080, anyone who expected one Radeon HD 6990M or GeForce GTX 580M to facilitate adequate performance is going to be sorely disappointed. One desktop card can certainly shoulder those loads, though, so we remain dissatisfied with the way mobile naming continues to be handled.

Information coming from a "reliable" source has informed DonanimHaber that AMD is expected to release its new Radeon HD 7900 series on January 9, 2010. This is a day ahead of the typical Tuesday release, along with one day before the official start of CES 2012. It is expected to have two cards released, the HD 7970 and HD 7950, that are based on AMD's new 28 nm "Tahiti" platform. Early specifications for "Tahiti" (based on bits and pieces of information from various sources) are listed below. These specifications match closely to what was originally reported in September for the HD 7970 and HD 7950. 

• 4.50 billion transistors, die-area of 380 mm², built on TSMC 28 nm process
• Advanced GCN 1D architecture
• 2048 1D processing cores
• 128 TMUs, 48 ROPs
• 384-bit wide GDDR5 memory interface, memory clock slightly below 1 GHz, target bandwidth of 240~264 GB/s

In addition, it is reported the new HD 7900 series will introduce Eyefinity 3D. This will add support for 3D-optimized (120 Hz) displays, and allow users to create large stereoscopic 3D display heads using multiple 3D displays. All of this looks to be coming with a hefty price though, as early European pricing for the HD 7970 is in the range of 539-549 Euro (or $699.03 to $712.00)

GeForce GTX 590: Bringing The Heat

Today, the worst-kept secret in technology officially gets the spotlight. Hot on the heels of AMD’s Radeon HD 6990 4 GB introduction three weeks ago, Nvidia is following up with its GeForce GTX 590 3 GB. According to Nvidia, it could have introduced this card more than a month ago. However, we know it continued revising its plans for a new flagship well into March. The result is a board deliberately intended to emphasize elegance, immediately after the Radeon HD 6990 bludgeoned us over the head with abrasive acoustics.

Pursuing quietness might sound ironic, given that GPUs based on Nvidia’s Fermi architecture are notoriously hot and power-hungry. To think the company could put two on a single PCB and not out-scream AMD’s dual-Cayman-based card is almost ludicrous. And yet, that’s what Nvidia says it did.

It admits that getting there wasn’t an easy task, though. Compromises were made. For example, Nvidia uses the same mid-mounted fan design for which we chided AMD. It dropped the clocks on its GPUs to help keep thermals under control. And the card still uses more power than any graphics product we’ve ever tested.

But it’s quiet. Crazy-freaking quiet. The quietest dual-GPU board I’ve tested since ATI’s Rage Fury Maxx (how’s that for back-in-the-day?). Mission accomplished on that front. The question remains, though: was Nvidia forced to give up the farm just to show AMD that hot cards don't have to make lots of noise?

Under The Hood: Dual GF110s, Both Uncut

In my discussions with Nvidia, the company made it clear that it wanted to use two GF110 processors, and it didn’t want to hack them up. Uncut GF110s, as you probably already know from reading GeForce GTX 580 And GF110: The Way Nvidia Meant It To Be Played, employ four Graphics Processing Clusters, each with four Streaming Multiprocessors. You’ll find 32 CUDA cores in each SM, totaling 512 cores per GPU. Each SM also offers four texturing units, yielding 64 across the entire chip. Of course, there’s one Polymorph engine per SM as well, though as we’ve seen in the past, Nvidia’s approach to parallelizing geometry doesn’t necessarily scale very well.

The GPU’s back-end features six ROP partitions, each capable of outputting eight 32-bit integer pixels at a time, adding up to 48 pixels per clock. An aggregate 384-bit memory bus is divisible into a sextet of 64-bit interfaces, and you’ll find 256 MB of GDDR5 memory at all six stops. That adds up to 1.5 GB of memory per GPU, which is how you arrive at the GeForce GTX 590’s 3 GB.

Nvidia ties GTX 590’s GF110 processors together using its own NF200 bridge, which takes a single 16-lane PCI Express 2.0 interface and multiplexes it out to two 16-lane paths—one for each GPU.

What changed from the ill-received GF100-based GeForce GTX 480 to GF110? From my GeForce GTX 580 review:

“The GPU itself is largely the same. This isn’t a GF100 to GF104 sort of change, where Shader Multiprocessors get reoriented to improve performance at mainstream price points (read: more texturing horsepower). The emphasis here remains compute muscle. Really, there are only two feature changes: full-speed FP16 filtering and improved Z-cull efficiency.

GF110 can perform FP16 texture filtering in one clock cycle (similar to GF104), while GF100 required two cycles. In texturing-limited applications, this speed-up may translate into performance gains. The culling improvements give GF110 an advantage in titles that suffer lots of overdraw, helping maximize available memory bandwidth. On a clock-for-clock basis, Nvidia claims these enhancements have up to a 14% impact (or so).”

Other than that, we’re still talking about two pieces of silicon manufactured on TSMC’s 40 nm node and composed of roughly 3 billion transistors each. At 520 square millimeters, GF110 is substantially larger than AMD’s Cayman processor, which measures 389 mm² and is made up of 2.64 billion transistors.

Now, it’s great to get all of those resources (times two) on GeForce GTX 590. However, while the GeForce GTX 580 employs a 772 MHz graphics clock and 1002 MHz memory clock, the GPUs on GTX 590 slow things down to 607 MHz and 853 MHz, respectively.

As a result, this card’s performance isn’t anywhere near what you’d expect from two of Nvidia’s fastest single-GPU flagships. That might be alright, though. After all, AMD launched Radeon HD 6970 as a GeForce GTX 570-contender; the 580 sat in a league of its own. So, although AMD’s Radeon HD 6990 comes very close to doubling the performance of the company’s quickest single-GPU cards, GeForce GTX 590 doesn’t have to do the same thing in order to be competitive at the $700 price point AMD already established and Nvidia plans to match.

We already know what AMD had to do in order to deliver “the fastest graphics card in the world.” Now, how does Nvidia counter?

Building A Dual-GPU Beast...And Keeping It Classy?

The Radeon HD 6990 4 GB is a card that makes no apologies. It’s big. It’s fast. It’s hot. It’s expensive. It’s loud. Some of the comments on my review of that board tried to draw parallels between the 6990 and Bugatti’s Veyron supercar. That’s way too complimentary. It’s really more of a Viper. The Viper is still sexy, but it’s nowhere near as masterfully crafted as a Bugatti, and AMD’s flagship doesn’t hesitate to get up in your face as it delivers its performance. Nvidia claims it was going more for Porsche-like nimbleness. Again, that’s hardly seems plausible given what we know sits under the GeForce GTX 590’s hood. But I didn’t jump to conclusions about AMD’s card until I had the data, and I won’t do that here, either.

Long, Strong, Something About Friction…

Nvidia’s card is big as well. Whereas the GeForce GTX 580 measures 10.5” long, the 590 is 11” long. But that’s a full inch shorter than AMD’s 12” Radeon HD 6990 (and indeed the 5970, which was also a 12” board).

Similar to the 6990, Nvidia employs a mid-mounted fan design. The company says this was necessary—the blower-style fan responsible for cooling the GeForce GTX 580 would have generated far too much noise as it tried to dissipate the board’s rated 365 W. Be that as it may, I didn’t like the challenges of pumping gratuitous heat into my chassis with AMD’s design, and it doesn’t get any more attractive coming from Nvidia. The difference is that, while AMD employed a mid-mounted fan and failed to strike a balance on the acoustic side, Nvidia gives us a more useable experience.

Perhaps surprisingly, the company doesn’t deviate too far from the game plan AMD executed. Its GF110 GPUs sit on either side of the center-mounted fan. Vapor chamber-based heatsinks cover each graphics processor, and an array of aluminum fins on each of them directs air out the front and back. But AMD uses a centrifugal fan. Nvidia employs an axial fan. And that distinction apparently makes the difference in noise output. Nvidia also claims that its vapor chamber design is superior, but without any way to quantify those claims or the benefit of AMD’s phase-change thermal interface material that supposedly helps improve thermal performance by 8%, we’ll stop short with the fan analysis.

As far as exhaust goes, Nvidia is actually at a disadvantage. AMD’s elegant use of DisplayPort allows the Radeon HD 6990 to offer five independent outputs, while still leaving an entire expansion slot for pushing heat out of the card’s shroud. Nvidia exposes three dual-link DVI outputs and one mini-DisplayPort connector—arguably a more klugey way to enable fewer outputs, but justifiable when you consider each GF110 is limited to two outputs anyway. Between those four ports, the 590 is left with little more than half of one slot for exhaust.

It’s also worth noting that Nvidia’s fan speed ramp is more granular than AMD’s. During testing, I found that the Radeon HD 6990 spins at 2180, 2880, or 3600 RPM. Stepping between those speeds is very obvious. The 590 adjusts more dynamically, and you really don’t notice when it accelerates. At the risk of giving too much away on page two, the result is one decibel more than the already-quiet GeForce GTX 580 under maximum load.

More Power Means More Heat

Nvidia’s exceptional acoustic performance is all the more spectacular when you consider its power requirements. According to the company, the GeForce GTX 590 has a 365 W thermal limit. That would seemingly compare favorably to AMD’s Radeon HD 6990, rated at 375 W at its stock 830 MHz and 450 W overclocked (and overvolted) using the built-in firmware selector. The thing is, on average, the GeForce GTX 590 uses more power than the 6990 at that overclocked setting. I mean, the two setups are close (separated by 5 W), but the 590 is still the highest-power offering, despite Nvidia’s rating.

Driving such a power-hungry card naturally means using two eight-pin PCI Express power connectors, and Nvidia recommends at least a 700 W PSU to get the job done.

Display Outputs And Tessellation Performance

As previously mentioned, GeForce GTX 590 comes equipped with three dual-link DVI outputs and one mini-DisplayPort connector. You can use all four concurrently in productivity-oriented environments, but taking advantage of Surround or 3D Vision Surround limits you to three. Be aware that creating a single display surface via Surround mode kills HDCP compliance for things like Blu-ray movie playback.

This is one comparison where AMD comes out on top definitively. We’ve seen the company enjoy much success with Eyefinity, and that halo feature remains the reason I continue using Radeon-based cards in my production workstation. Even with two GPUs on a single card, Nvidia’s flagship cannot match the output connectivity of a single Radeon HD 6000-based board. Hopefully this compels the company to devote more attention to what has become a must-have feature for the growing contingent of folks who’d rather spend a couple hundred bucks each on three 1920x1080 displays versus $1500+ on a nice 2560x1600 monitor.

Another Look At Tessellation

I’ve been fairly critical of Nvidia’s scaling story when it comes to geometry performance. Titles like HAWX 2 have proven to be demanding enough to demonstrate the relative weakness of AMD’s Radeon HD 5000-series cards. However, I haven’t been able to show the real benefit of moving from eight Polymorph engines (in the GeForce GTX 560 Ti) to 15 or 16 (in the GeForce GTX 570 and 580).

Meanwhile, AMD’s Radeon HD 6990 showed us that four of the company’s own tessellation engines can maintain 76% of the card’s performance with the feature turned on versus disabling it completely.

The GeForce GTX 590 reestablishes Nvidia’s dominance by achieving 83% scaling, shedding just 41 frames when tessellation is turned on at 1920x1200. It probably bears further examination as to what, exactly, prevents better scaling on the single-GPU cards. But I don’t think the problem is related to geometry throughput.

Test Setup And Benchmarks

Benchmark Results: 3DMark 11

We’ve yet to see Nvidia perform well in 3DMark 11. The company claims that this title doesn’t lean heavily enough on tessellation, which it thinks would improve its standing.

As I’ve said before and will repeat again, 3DMark is a synthetic metric that attempts to factor out the political jockeying that affects the outcome of almost every other game out there. Nvidia believes it emphasizes lighting too much, AMD is under the impression it includes too much tessellation, and Intel is mad that you need DirectX 11 to run it at all.

That’s not to say the GeForce GTX 590 does poorly. It falls in just behind the Radeon HD 6990 at its stock 830 MHz, easily besting the Radeon HD 5970. The end result isn’t bad, but with both AMD’s and Nvidia’s dual-GPU solutions priced identically at $700, the 6990 starts off with an advantage. Now, let’s see if the card is able to hold its lead.

Benchmark Results: Metro 2033 (DX11)

The GeForce GTX 590 flexes its muscle in Metro 2033, particularly with no anti-aliasing or anisotropic filtering enabled. It’s even playable-ish at up to 2560x1600. Why are these frame rates so low? We have the debilitating DirectCompute-based depth-of-field filter enabled, which essentially halves frame rates. If you turn that off, all of these numbers jump substantially.

We see the GeForce GTX 590 stumble in one situation: 2560x1600 with AA and AF turned on. The assumption here is that 1.5 GB of graphics memory isn’t enough for such taxing options, given the Radeon HD 6990’s relative success, the GeForce GTX 580’s predictable finish, and the Radeon HD 5970’s outright failure to turn back reasonable frame rates with 1 GB of memory per Cypress processor.

The only problem with running out of steam at the top end is that’s where this card is intended to dominate. We wouldn’t have thought this setting playable anyway, given the 39 FPS results without AA/AF; however, that doesn’t necessarily bode well for quad-GPU comparisons between GeForce GTX 590 and Radeon HD 6990.

Nevertheless, at 1920x1080, the GTX 590 takes first place and serves up amply-quick results to qualify as playable.

Benchmark Results: Lost Planet 2 (DX11)

The dual-GF110-equipped GeForce GTX 590 continues its campaign in Lost Planet 2, outgunning the Radeon HD 6990 even in its overclocked, warranty-voiding mode. And should we be surprised? This is another title classified under the The Way It’s Meant To Be Played program.

Are we saying this necessarily predisposes it to foul play? No. At the same time, it’s probable that Capcom’s team spent more time with Nvidia hardware than the competition, so it's at least better-optimized for GeForce cards right out the gate. I have to imagine that Nvidia would argue this is a positive side-effect of its development engagement. Folks with Radeon cards likely feel different.

While AMD used to fare much worse in this one, the company’s Radeon cards perform much better now after a number of Catalyst driver updates.

All of that aside, it’s interesting to note that Nvidia suffers the same loss at 2560x1600 with anti-aliasing turned on. The difference here is that we’re looking at frame rates in the 60s—plenty playable. Losing at a viable setting stings more, since we’d expect the GTX 590 to fall in the 70 FPS range, given scaling at other resolutions.

Benchmark Results: Aliens Vs. Predator (DX11)

When it comes time to pit flagship against flagship, the GeForce GTX 590 gets rocked in Aliens vs. Predator. Both high-end boards deliver more than ample performance at 1680x1050 and 1920x1080. It’s 2560x1600 where you really need one of these cards for high-end frame rates (especially with anti-aliasing turned on). Regardless of the comparison you draw, though, AMD’s Radeon HD 6990 takes a first-place finish in this title. Incidentally, AvP was one of the games AMD's team spent extra time helping optimize for Radeon boards prior to launch.

Stepping down to the fastest single-GPU board puts you under 40 FPS, even without AA, and it’s all downhill from there.

Benchmark Results: Battlefield: Bad Company 2 (DX11)

We see the closest finish yet in Battlefield: Bad Company 2. Both Nvidia’s GeForce GTX 590 and AMD’s Radeon HD 6990 perform almost identically. At 1680x1050, the GeForce card is on top, if only barely. By the time you hit 2560x1600, though, it’s putting down numbers slightly behind the 6990 at its stock settings.

If you’re playing on a 30” display and cranking up the visual details, a dual-GPU configuration is really the way to go in Battlefield. The fastest single-GPU card ducks in under 60 FPS, while the GTX 590 and 6990 hover above 80 FPS.

Benchmark Results: F1 2010 (DX11)

We’ve already seen a couple of titles endorsed by Nvidia’s marketing department, and a couple of others influenced by AMD’s software development team. F1 2010 is a third that goes AMD’s way. In fact, even the Radeon HD 5970 is faster than the GTX 590 in F1 2010.

At 1680x1050, we see the clearest sign of a processor bottleneck encountered yet. At 1920x1080, the top three configurations are still relatively jammed up, even with 8x AA enabled. There’s more range between each card by the time we hit 2560x1600. But the fact that the Radeon HD 5970, with its 1 GB of memory per GPU, has no trouble at all outmaneuvering the brand new GeForce GTX 590 at our most demanding settings makes it pretty clear that this title’s bottleneck is elsewhere.

Benchmark Results: Just Cause 2 (DX11)

Just Cause 2 might be a TWIMTBP title, but AMD’s Radeon HD 6990 flips the GeForce GTX 590 on its head here. The dual-Cayman card takes a victory at all three resolutions, outmaneuvering Nvidia’s dual-GF110 contender. AMD’s Radeon HD 5970 even shows off a bit at 1680x1050 and 1920x1080. Once we hit 2560x1600, though, the GeForce GTX 590 slots under the 6990. Performance with anti-aliasing and anisotropic filtering is much closer, and the cards underneath Nvidia’s new flagship fall off a lot faster with these demanding settings dialed in.

Benchmark Results: World Of Warcraft (DX9)

Nvidia’s cards tend to perform better in World of Warcraft, consistent with our Cataclysm Performance Guide. Most of the differentiation happens when you enable anti-aliasing, though.

Using 8x MSAA, the GeForce GTX 590 does score a first-place finish at 2560x1600. However, the Radeon HD 6990 manages to achieve second place, despite its position in the chart (sorted according to the non-AA scores, just like every other game title). The GeForce GTX 580 comes in third, followed by AMD’s Radeon HD 5970.

Quad-SLI: Something You Need To Plan Out

I'm pretty gosh-darned convinced that the most sensible application for dual-GPU boards is now quad-GPU configurations. If you're not running four GPUs, put a couple of single-GPU cards in CrossFire or SLI, spend less money, and avoid dumping gratuitous heat into your chassis. It's as simple as that.

Should you decide to take the plunge, though, and pursue unbridled performance via four graphics processors operating concurrently, there are precautions that need to be taken (especially when you're talking about these specific designs with mid-mounted fans).

Ahead of the Radeon HD 6990 launch, AMD gave no indication of what it'd take to properly support a pair of Antilles boards in four-way CrossFire. Even now, you can hit the 6990's product page and find very little detail on what you need to enjoy a trouble-free experience. That is a mistake, in my opinion, because you cannot drop 750 W worth of graphics cards into any platform with the slots to spare and expect them to run well. You need the right power supply, the right motherboard, and most of all, the right enclosure. Certain cases simply cannot cope with volume of heated air that gets recirculated.

Nvidia addresses the need for more information with its own list of validated components. And while we don't necessarily like the limited number of options resident on those lists (especially the very short group of chassis that get a nod), I'm more comfortable with slim pickings than a dice roll.

If you plan to build using two GeForce GTX 590s, here are the lists of approved components as they exist thus far:

Quad-Approved Motherboards

Perhaps the most important quality to look for in a motherboard is proper slot spacing. Whether you're in the market for two GTX 590s or a pair of HD 6990s, there needs to be at least one vacant slot worth of space between the cards. The following list is composed of platforms approved by Nvidia to support dual GeForce GTX 590s.

All of these models might work well thermally, mechanically, and acoustically, but I'd also want to conduct extra tests to determine if splitting 16 lanes of aggregate PCI Express connectivity between four GPUs on those P67- and P55-based boards yields acceptable scaling. All of our single-card testing was done on an X58 board. However, the system we used for benchmarking SLI'd and CrossFire'd configs centers on P67.

Quad-Approved Power Supplies

The principle concern when it comes to power delivery is making sure you have enough sustainable output. Nvidia's list of recommendations incorporates 1100+ W models with at least four eight-pin auxiliary power connectors.

Quad-Approved Chassis

Obviously, ventilation and physical orientation are both factors that have to be addressed in case choice. There are a lot of really nice-looking enclosures out there simply unable to cope with two 365 or 375 W graphics cards recirculating half of their dissipated heat. Currently, the list of validated enclosures is dismally-short. Hope you like one of these three options!

Of course, we really shouldn't be complaining, as we still haven't seen any recommendations from AMD. 

Assuming the DiY market for quad-GPU-based setups is miniscule, this is really more guidance for system builders than anything. We've been in contact with a couple different boutique folks through the past two launches, just trying to get a sense for what they think about working around these super-hot dual-GPU offerings. This time, we got extra lucky and scored the fruits of one company's pre-launch efforts...

Performance: Quad-SLI Versus Quad-CrossFire

Just prior to the launch, Maingear sent over its gorgeous Shift, all decked out with an overclocked Core i7-2600K running at 4.5 GHz. It included two GeForce GTX 590s and one Radeon HD 6990 (we already had one here in the lab to add to the system), allowing us to run benchmarks in an actual case and compare quad-GPU configurations from both AMD and Nvidia.

Maingear helped assuage one of my biggest concerns with its machine: mainly, the need for ventilation would force enthusiasts to adopt ugly, airy enclosures just to let the cards inside breathe. The Shift demonstrates that it's still possible to get a classy, monolithic PC (vertically-oriented, even) and still support GeForce GTX 590s and Radeon HD 6990 cards in quad-GPU configurations. The system does radiate a lot of heat under load, particularly toward the bottom, where these cards push a lot of air, but it handles the challenging thermals without a hiccup.

Most impressive is the fact that both configurations work without generating a ton of noise. Do not compare the following numbers to the results on page 17. Instead of measuring from 28" behind each card's exhaust port, we buttoned up the Maingear machine, stuck it under a desk three feet away, and noted the acoustic levels two feet off of the ground.

The AMD cards are still louder than Nvidia's in a closed system versus an open test bench. However, we're pleased to see Maingear's beefed-up cooling subsystem keeping air moving briskly enough to avoid the 3600 RPM fan setting.

Now we can start comparing the performance of two Radeon HD 6990s to a pair of GeForce GTX 590s. Just as we saw in the single-card results, Nvidia's flagship part dominates, so long as you don't exceed the capabilities of its onboard memory. With 4x MSAA applied, AMD's Radeon HD 6990 in one- and two-card configurations is the faster solution. And you really need two of those cards (four Cayman GPUs total) to make Metro 2033 playable at 2560x1600 using our demanding settings.

Although AMD gets the upper-hand in the most taxing environments, we get the sense that its drivers aren't yet as ready for quad-GPU operation, despite the almost three-week head start its team has on Nvidia's launch. Two 6990s in CrossFire aren't even able to start the Lost Planet 2 benchmark before crashing to the Windows desktop.

AMD's problems continue in Aliens vs. Predator, where even its newest Catalyst 11.4 preview driver doesn't seem to serve up scaling between two cards. As a result, where we see one Radeon HD 6990 outperform the GeForce GTX 590, two 590s take a commanding  lead.

In single-card mode, AMD's Radeon HD 6990 is only slightly faster than the GeForce GTX 590. But adding a second card shows that four Cayman GPUs scale better than a quartet of GF110s in Battlefield: Bad Company 2.

One Radeon HD 6990 is faster than one GeForce GTX 590 in Just Cause 2. Adding a second card allows AMD to maintain that lead without AA/AF. However, the GeForce GTX 590s in SLI manage to pass AMD's solution once anti-aliasing is applied. The performance is close, just as we'd expect given identical price tags.

Multi-Monitor: GeForce GTX 590 Vs. Radeon HD 6990 At 5760x1080

Another set of requests from the Radeon HD 6990 review asked us to set aside the single-display configurations for a moment and spend some time testing multi-monitor setups. That might actually become something we do regularly, given the high price of 30” displays compared to smaller 1920x1080 screens. So long as the contenders are all triple-screen-capable, of course…

Without anti-aliasing and anisotropic filtering applied, the GeForce GTX 590 scores a narrow victory in Metro 2033 at 5760x1080. However, my postulations about available graphics memory seem to be coming true here. Even more so than at 2560x1600, 1.5 GB per card is just not enough to facilitate this resolution at aggressive settings, resulting in a veritable slide show from Nvidia’s newest card. And it’s not that the 6990 is playable. But its 2 GB per GPU is able to cope much better.

Nvidia’s card takes the lead without AA turned on in Lost Planet 2. Enabling it, though, exacts a performance penalty that gives AMD’s Radeon HD 6990 a victory.

The same thing happens in Aliens Vs. Predator. Nvidia finishes on top in the easier workload, but gives up ground under the stress of anti-aliasing.

Bad Company 2 throws us for a loop, seemingly reversing the trend observed in the three preceding charts. Here, it looks like AMD is on top without AA, and then the GTX 590 kicks into gear with 8xAA/16xAF enabled.

The scores are attributable to a bug in Nvidia’s driver, though. Without AA/AF, the 590 renders the scene correctly. With the eye candy turned on…well, see for yourself.

Nvidia takes first place in Just Cause 2 with and without AA/AF enabled.

Dual-GPU Cards Versus Two Cards In SLI/CrossFire

We had several requests for GeForce GTX 580 cards tested in SLI after the Radeon HD 6990 review. So, in addition to factoring in results for the GeForce GTX 590, we’re also including scores for two 580s.

It takes nearly $1000 worth of GeForce GTX 580s to beat a $710 Radeon HD 6990 in 3DMark 11. That alone doesn’t mean much, given this title’s synthetic nature. More interesting is the fact that two Radeon HD 6970s in CrossFire, which sell for $680 together, turn in faster scores. GeForce GTX 570s also do really well, and you can find a pair of those for $650. In comparison, the GeForce GTX 590 falls into last place, making it both more expensive and slower than a pair of its own derivative boards.

We noticed that the 1.5 GB of memory on Nvidia’s highest-end cards seemed disadvantaged at 2560x1600 in Metro 2033 with anti-aliasing and anisotropic filtering cranked up. So, while the GeForce GTX 580 is followed by the 590, followed by the 570 without the “extra” eye candy turned on, all three cards trail the three AMD-based configurations when the knobs go to 11.

Two GeForce GTX 580s serve up the ultimate performance in Lost Planet 2, and that setup dominates, regardless of whether AA is enabled or not. Again, though, the GeForce GTX 590 starts falling behind once the anti-aliasing gets cranked up. All three AMD-based configurations slide on by, if only by a couple of frames per second.

AMD’s sponsorship of this game can’t keep a pair of GeForce GTX 580s from taking first place. Three Radeon-based configs take second through fourth place, though. The GeForce GTX 590 has to settle for a last-place finish here—not a great place for a $700 card to land.

Nvidia’s GeForce GTX 590 ties the Radeon HD 6990 for last place, but AMD’s card shows up ahead because it serves favorable performance with AA and AF enabled. The overclocked Radeon HD 6990 and GeForce GTX 570s in SLI go head-to-head, but are both outmoded by two Radeon HD 6970s in CrossFire. Even that solution gets dominated by two GeForce GTX 580s in SLI, though.

Just Cause 2 is all about AMD’s hardware. The three Radeon-based setups take first through third place (led by the two 6970s in CrossFire), while Nvidia’s GeForce GTX 580s in SLI place fourth. The company’s new flagship settles for a fifth-place finish.

Power, Noise, And Air Temperatures

I don’t think this one will surprise anyone. At idle, the GeForce GTX 590 draws more power than any other graphics card in our comparison, taking that un-honor away from the GeForce GTX 480.

It’s probably fair to say that enthusiasts spending $700 on graphics aren’t really concerned by the 32 W spread between Nvidia’s new flagship and the lowest-power option in this chart, though. Let’s see if the situation gets any direr under load.

Nvidia’s GF110 is a more power-hungry GPU than AMD’s Cayman. However, Nvidia still faces very real limits with two graphics processors on a PCB. Those constraints force the company to drop the clocks on GeForce GTX 590 in a significant way. The result is a power consumption chart that looks very much like AMD’s overclocked, voided-warranty Radeon HD 6990 at 880 MHz.

The obvious discrepancy there is that AMD rates that particular configuration for 450 W, while Nvidia says the GeForce GTX 590 is a 365 W board. Average out those lines, and you’ll see the following system power use:

The GeForce GTX 590 is actually more power-hungry than the overclocked Radeon HD 6990. It also makes the GeForce GTX 480 look like an econobox, which of course it’s not.

From a power bill-perspective, that’s scary stuff. But again, you probably don’t care, right? It’s a difference of 5 W up against the Radeon card. More worrying is what that power means in terms of heat dissipation, and consequently, noise. AMD doesn’t cope with the heat it generates well at all. Does Nvidia do any better?

Dear goodness, yes. The GeForce GTX 590 is like, 1 dB louder than the GeForce GTX 580. How can this be? "Surely Nvidia must be doing something insidious to slow the 590 down under extreme duress to keep from overwhelming the smaller card’s cooler," the fan boys will wail. But no—I logged clocks and temps (fan speeds wouldn’t register) in GPU-Z, and found that the GeForce GTX 590 ran at 607 MHz, never exceeding 87 degrees or so. This is, absolutely, where Nvidia stands head and shoulders above what AMD achieved with its Radeon HD 6990.

Now, I’ve already presented much of this data in my Radeon HD 6990 review, and judging by the comments, there are those out there who simply don’t want to believe that the 6990 is so loud as to be disruptive. So, here you go; hear for yourself. Both videos were taken with the Sennheiser wireless mic at 00 dB sensitivity, 28” back from the I/O bracket of each card. Listen to one, and then the other (particularly at the end of each video—that’s where AMD’s card heats up and gets the loudest).

The standard answer contesting my noise complaints in the 6990 story was “any enthusiast willing to spend $700 on a 6990 should expect to pay for water cooling, too.” I reached out to the folks at Danger Den to find out when water blocks might be available for either of these cards. The Nvidia block will be available when the 590 launches, and the AMD block should emerge the following week. Prices are expected to land around $120 for both. If that’s the route you want to go, fine. But you’re getting awful close to a pair of GTX 580s at that point—and two 580s are going to be significantly faster than either dual-GPU flagship.

Because Nvidia enables three dual-link DVI ports on the front of its card, there’s limited room to push air from the front I/O bracket. AMD’s design is more elegant in this regard. I set forth to measure the air temps at both ends of each board using an Extech TM200 Dual K Thermometer.

Indeed, the temps at the front and back of AMD’s Radeon HD 6990 are more balanced, separated by a 10 degree difference. Nvidia’s GeForce GTX 590 also pumps out air heated to about 138 degrees Fahrenheit from the front, but air coming out the back is about 12 degrees warmer. I wish it were possible for me to measure air volume, in order to explore just how much air is getting recirculated with just one of these cards installed.

Justification for the mid-mounted fan is likely going to be the same for both AMD and Nvidia. Nvidia defends its choice, claiming it could have gone with a blower-style design that would have created an unacceptable acoustic experience. What we end up with at the end of the day isn’t ideal (especially for folks considering quad-GPU solutions and trying to pick a free-flowing-enough chassis), but it’s apparently the best way to keep noise at bay.

Power In SLI/CrossFire And Watts Per Frame

One chart that was missing from the Radeon HD 6990 story, and subsequently requested, was a power consumption chart comparing the dual-Cayman board to CrossFire’d and SLI’d configurations. “Fair enough,” I thought. “We’re telling folks to skip the 6990 entirely and buy 6970s or GTX 570s—it’s reasonable to want to know how those setups size up with regards to power use.” So, here’s that graph:

In addition to the 6990, 6970s, and 570s, I’ve also added the 590 and a pair of 580s in SLI.

Clearly, two GeForce GTX 580s ratchet up power use in a way that no other setup comes close to rivaling. The other solutions are much more evenly matched, though.

On average, the GTX 580s impose a 559 W system power figure running Metro 2033. Two Radeon HD 6970s in CrossFire are the next most power-hungry, averaging 442 W system power. The GeForce GTX 590-based system is just five watts behind, while two GTX 570s turn in an average power figure of 420 W. AMD’s Radeon HD 6990 (running at its stock clocks) winds up being the most power-friendly option, registering 402 W average system power.

Here’s the average frame rate each configuration achieved while we logged those power numbers.

A couple of folks looking to deflect criticism of the 6990’s value asked for a chart depicting watts per frame. There you go, gentlemen. The Radeon HD 6990 uses less power per frame per second than two Radeon HD 6970s, and both AMD-based configurations consume less system power per frame than all three Nvidia-based setups. Now, flip back to the previous page, replay the 6990 video, and think about all of the conversations you can enjoy not hearing!

Conclusion

Earlier this month, AMD established the Radeon HD 6990 as the fastest card on the planet, unseating its own Radeon HD 5970, the previous title-holder. Today, Nvidia challenges that crown, but ultimately fails to win consistently enough to unseat AMD’s flagship board in a meaningful way. The GeForce GTX 590 simply sacrifices too much in the way of clock rate for its 1024 CUDA cores to definitively overwhelm two Cayman GPUs operating in tandem. This was, of course, a necessary evil in order to fit a pair of unshorn GF110s into a reasonable thermal envelope. In contrast, AMD’s less power-hungry processors didn’t have to be as hamstrung.

Nevertheless, in a comparison between GeForce GTX 590 versus Radeon HD 6990, Nvidia wins.

And it wins because I can put a GeForce GTX 590 in my workstation, which doubles as my gaming machine, and not hear it. It’s quiet enough to use. And that’s a requisite.

“But wait Chris,” the fan boys cry out, using more vulgarity and fewer conventions of the English language. “Anyone with the money to buy a $710 Radeon HD 6990 will assuredly buy a water block. Does that not shift the win back in AMD’s favor?”

Well, there aren’t any water blocks available for the 6990 yet, but Jeremy at Danger Den was kind enough to let me know that he’ll have a solution available soon, priced around $120. So, add that to the price of the 6990 to solve its acoustic issues. Then factor in the cost of the rest of a water cooling circuit. Might as well buy a pair of GTX 580s at that point—they’re faster, after all.

The only way to really pit the newest dual-GPU cards against each other is in an arrangement that no single-GPU board can match, and that’s a quad-processor config. Water cooling really is the best way to go in that case, given intense thermal demands. However, Maingear showed us that you don't need water cooling to get the job done. Both the GeForce GTX 590 and Radeon HD 6990, in quad-GPU setups, are incredibly fast. Enthusiasts with generous credit limits and the desire for unprecedented performance can make a case to go with either $1400 setup. AMD doesn't shed performance as fast at 2560x1600 with details maxed out, but its drivers appear to be much less mature. Nvidia enjoys a higher degree of consistency, but again, that includes performance consistently falling off once the 1.5 GB per GPU gets overwhelmed.

If you’re only concerned with the performance of an individual dual-GPU card, then I happily repeat the recommendation I made in the 6990 review: buy two Radeon HD 6970s or two GeForce GTX 570s and enjoy all of the performance without having to balance thermals and acoustics. Unless you’re 1) worried about plugging up four expansion slots, 2) you really need the 6990’s fifth display output, or 3) the GTX 590’s sweet light-up logo won you over, don’t worry about which company is selling the “fastest graphics card in the world.” Rather, take advantage of the fantastic CrossFire and SLI scaling we now see from AMD and Nvidia (respectively), and save yourself a few bucks in the process. 

Follow @chris_angelini on Twitter

1985: The Founding

We want to commemorate the 10 years anniversary of the Radeon series with this little excursion into the past. We also want to pay our respects to ATI, whose name, following the acquisition by AMD and the launch of the Radeon HD 6000-series, is finally being removed from the market. After a quarter-century of ups and downs, hailed by fans and mocked by opponents, the end of the ATI brand marks the end of a company that has decisively influenced the history of graphics cards.

We would like to look back, once again, and mark the passage of years with ATI’s most important products. Of course, we can hardly imagine all of the company's products, so we're focusing on the desktop boards.

Let's begin in 1987...

1987 • EGA Wonder

1987 • EGA Wonder

The first-ever ATI graphics card, exclusively outputting EGA graphics.

1988 • VGA Wonder

1988 • VGA Wonder

A year later, ATI launched this 16-bit ISA card, run in an 8-bit ISA slot. It had an analog and a digital video connector so that all different kinds of available PC monitors could be connected.

1991 • Mach8 / Mach32

1991 • Mach8 / Mach32

ATI launched the Mach8 in the year 1991, which first was a co-processor, and then launched as an integrated processor in the shape of the Mach32. Thus, the year 1992 marks the year of the first 2D accelerator from ATI.

1994 • Mach 64 / Rage I

1994 • Mach 64 / Rage I

ATI launched the Mach64, its first 64-bit processor, in 1994. These cards already had features like color conversion, and were quite popular as powerful 2D accelerators. The card was later equipped with a 3D processor in the year 1995 and reappeared on the market as the Rage I.

1995 • New Name, New Logo

1995 • New Name, New Logo

The introduction of the Rage 3D chips was accompanied by a new logo on the boxes.

1996 • Rage II

1996 • Rage II

Great 2D performance (thanks to the old Mach64), but modest 3D performance was available on the Rage II. This card was widespread on the market, though often cursed because of driver issues.

1997 • Rage Pro

1997 • Rage Pro

ATI was able to somewhat catch up to the competition with its Rage Pro. The Rage IIc was almost as fast as Nvidia's Riva 128, making the relatively cheap card very popular.

1997 • Rage XL

1997 • Rage XL

The XL was nothing but a brutally-stripped Rage Pro aimed at the OEM market.

1998 • Rage 128

1998 • Rage 128

The lack of any kind of texture compression was not doing this card any favors. And then there were the driver issues, greatly infuriating users. This was the beginning of the end for ATI's Rage lineup.

1999 • Rage Fury Maxx

1999 • Rage Fury Maxx

This was the first dual-GPU card from ATI. Unfortunately, two chips on a PCB (along with two times the frame buffer) did not do much to hide the shortcomings of the design. Now a popular collector’s item, it was a bad investment for many enthusiasts at the time.

2000 • A New Era: Radeon 8000

2000 • A New Era: Radeon 8000

Graphics cards were sold under the new brand name Radeon for the first time in 2000; the name Rage was probably too damaged. New name, new luck? Let us recall…

2000 • Radeon 7500

2000 • Radeon 7500

So its performance did not exactly blow peoples’ minds. But the first Radeon was a solid entry.

2001 • Radeon 8500

2001 • Radeon 8500

Reasonable value for money earned the Radeon a new following. The LE models were entry-level, whereas the XT models were high-end.

2002 • Multimedia On-Board: Radeon 8500 All-In-Wonder

2002 • Radeon 8500 All-In-Wonder

ATI combined a 3D graphics card with an analog TV and radio tuner and made them available on its All-In-Wonder. This was a successful step into an OEM market where Nvidia was never able to succeed.

2002 • Radeon 9000: The Entry-Level GPU

2002 • Radeon 9000: The Entry-Level GPU

The Radeon 9000-series launched in 2002 with a entry-level model: the 9000.

2003 • Radeon 9600 Pro: The Popular One

2003 • Radeon 9600 Pro: The Popular One

After a few weak 9000-series models, the Radeon 9600 Pro arose to become the most popular mid-range model at the time. It was a good card with solid frame rates, except for a lack of Shader Model 3.0 support.

2003 • Radeon 9800 XT: ATI’s Hot Iron

2003 • Radeon 9800 XT: ATI’s Hot Iron

Nicknamed “Comet,” it was extremely fast, not especially quiet, but above all: at risk for heat exhaustion. We buried two burnt-up 9800 XTs ourselves, trying to cope with the (financial) loss. Going out in flames is just not cheap.

2004 • Radeon X700: The Xs Appear

2004 • Radeon X700: The Xs Appear

ATI created a whole new volume upper-mid segment model with the X700. Though affordably priced, the Shader Model 2.0b design restriction negatively impacted its competition with Nvidia's GeForce 6-series. The Nvidia cards offered more 3D effects with the help of Shader Model 3.0, and, to make things even more interesting, resourceful users could quite effortlessly unlock the GeForce 6600 card to a GeFore 6800 (at least initially).

2005 • Radeon X850 XT: The Limited-Functionality Burner

2005 • Radeon X850 XT: The Limited-Functionality Burner

ATI’s answer to the Nvidia FX 5800 Ultra "Hair Dryer:" the Radeon X850 XT.

The card was extremely fast, and superior to the GeForce 6800 GT in many ways, so long as Shader Model 2.0b was enough. The situation was much closer once you took Shader Model 3.0 into account, and many games would not start at all on the AMD hardware. Did you want to run Bioshock with an X850 XT? You had to wait (a long time) for the community to produce a hack.

2005 • X1300 Pro: A Slow Start

2005 • X1300 Pro: A Slow Start

E.T. looked just as sad on this Sapphire card as most customers who tried playing games on it. Sure, it had DirectX 9 and Shader Model 3.0 support for nice graphics, but the performance just wasn’t there.

2006 • X1650 Pro: The Mediocre Mid-Range

2006 • X1650 Pro: The Mediocre Mid-Range

ATI tried to do battle with the seventh-generation GeForce cards using its Radeon X1650 Pro, even if only in the mid-range segment.

2006 • X1950 XTX: The Late High Flyer

2006 • X1950 XTX: The Late High Flyer

Out of the way! It's late, but here it comes anyway: the fat Big Block from ATI. Using the X1950 XTX, ATI swept away all GeForce 7-series cards from Nvidia and grabbed the performance crown. It was popular and extremely fast, but also hot and expensive.

Altogether, the Radeon X1950 XTX is the first thing that comes to mind when trying to remember the times ATI has beaten its competitor (Ed.: I don't know about that; I always think back to R300, personally).

The GeForce 8800 was the first card able to stand up to it, but that was an impressive card in itself. And thus began the history of graphics cards with unified shader architectures.

2007 • HD 2600 Pro: Unified Shader, Light Edition

2007 • HD 2600 Pro: Unified Shader, Light Edition

The GeForce 88xx series did not face much opposition from AMD's high-volume Radeon HD 2600 Pro. Even the 8600 GTS did not have much to fear, really.

However, the ATI card did offer the possibility of outputting audio directly through the DVI output for the first time. Not a bad argument for the first HTPCs. Manufacturers like HIS set new standards with powerful and quiet cooling solutions.

2007 • HD 2900 XT: The Blown GeForce Hunter

2007 • HD 2900 XT: The Blown GeForce Hunter

With brutal raw performance, the fanboys were not the only ones surprised that this card hopelessly lagged behind when it was launched. Only much later, with better drivers, was it able to make up ground. It was loud, hot and in the end not overly successful.

2007 • HD 3850: The Third Attempt

2007 • HD 3850: The Third Attempt

ATI did not exactly create a powerhouse with the Radeon HD 3850, but it was still a good all-around card. Speed-wise, it stood no chance against competing GeForce cards. But a reasonable price tag made it a popular mid-range volume product.

2007 • HD 3850 AGP: Niche Solution Instead Of Peak Power

2007 • HD 3850 AGP: Niche Solution Instead Of Peak Power

Take a bridge chip and make a modern PCIe card AGP-compatible again. It was a great concept, but suffered from mediocre drivers. It was never able to take advantage of the latest Catalyst drivers, so customers either went with the AGP competition or gave up on the interface altogether.

2008 • HD 3870 X2: Tag Team, Back Again

2008 • HD 3870 X2: Tag Team, Back Again

It took a very serious card to even attempt breaking through Nvidia's GeForce dominance at this point. After all, by now, G80 had turned into an even more efficient G92.

Fortunately for ATI, the Radeon HD 3870 X2 was enough to take back the performance crown, even if it took a considerate amount of silicon to address the issue. Power consumption was not this board's strong suit, needless to say.

2008 • HD 3850 X2: An Exotic From Asus

2008 • HD 3850 X2: An Exotic From Asus

Asus designed its own card with two Radeon HD 3850 chips, but failed to attract much interest from buyers.

2008 • HD 4670: The Mid-Range, Redefined

2008 • HD 4670: The Mid-Range, Redefined

Without any additional power connectors, the low-power HD 4760 was an interesting card from ATI adopted by the HTPC crowd. Even Assassin’s Creed ran quite fluidly at 1680x1050.

2008 • HD 4870: Extremely Popular, Just As Affordable, And At Least As Hot

2008 • HD 4870: Extremely Popular, Just As Affordable, And At Least As Hot

With its Radeon HD 4870, ATI had done it again, approaching the performance crown. It overtook Nvidia's competing cards in a few benchmarks, but the bigger news was that the company was back.

A slightly lower price made up for any shortcomings, such as higher idle power consumption and heat generation.

2008 • HD 4870 X2: Another Double Decker

2008 • HD 4870 X2: Another Double Decker

Woah! If size was everything, this double decker would be a star (at least when the drivers enabled CrossFire). The joy of owning this card was completely dependent on software, it seemed.

2009 • HD 4890

2009 • HD 4890

Take a success story, squeeze it just a bit more and…no, we're not talking about the G92 this time. Rather, we're looking at the Radeon HD 4890, based on the same general architecture at the heart of Radeon HD 4870.

2009 • HD 5770: Middle-Class With Class

2009 • HD 5770: Middle-Class With Class

ATI managed to make the Juniper family an attractive mid-range option. Low idle power consumption, moderate power consumption under load, and low production costs (complements of what the company called its Sweet Spot strategy) meant economical gains for ATI (already part of AMD by this point).

2009 • HD 5870: A Hot Autumn Ends With A Cool High-End

2009 • HD 5870: A Hot Autumn Ends With A Cool High-End

A November child of the other kind, ATI's Radeon HD 5870 was a rather affordable high end-card that Nvidia only outpaced with the dual-GPU GeForce GTX 295.

2009 • HD 5970: A Tasty Sandwich Without The Double X

2009 • HD 5970: A Tasty Sandwich Without The Double X

Two can play at that game. ATI re-reclaimed the performance crown yet again with a dual-GPU Radeon HD 5000-series board. The computing power of a single HD 5970 is all it takes to get to the top ten in the protein folding charts.

2010 • HD 5670: From Top To Bottom

2010 • HD 5670: From Top To Bottom

The new lower mid-range segment? ATI now has so many variations in its portfolio that the classes merge into each other. So long as the end-users are happy, though, and can keep track of what is what, then cards like the 5670 will continue to sell.

2010 • The End Of The ATI Brand

2010 • The End Of The ATI Brand

As the year came to an end, only the Radeon brand remained standing (even the Stream initiative took a bullet).

After 25 years, ATI finally becomes AMD. Though it's a notable passing, the move makes sense for AMD. The company that once swallowed successful competitors like Tseng Labs and ArtX is now completely assimilated. Good night ATI; good morning AMD!

2010 • HD 6850: The x800s Become Middle-Class

2010 • HD 6850: The x800s Become Middle-Class

Anyone who thinks that the new 6850 and 6870 are the new upper-class, as AMD's past naming convention would suggest, is badly mistaken. AMD defined a new middle class with these products, while the Radeon HD 5700-series continues cleaning up in the lower end of the mid-range segment.

2010 • HD 6870: The Big Small One

2010 • HD 6870: The Big Small One

The Radeon HD 6870 introduced solid performance at a reasonable price, with new features like up to four display outputs per card.

Of course, as we recently saw, the Radeon HD 6900-series cards followed up by making AMD's architecture more efficient. And the Radeon HD 6990 should be here soon to demonstrate what a dual-GPU version of the Cayman family can do.

The Other Green Alternative

Some say green is the new red, and no place is that more true than in the rebranding of ATI to AMD. And yet, marketing similarities between AMD and Nvidia don’t end in color, as the newer, more efficient Radeon HD 6850 offers performance and a price nearly identical to Nvidia's GeForce GTX 460 1 GB. If you want to know more about the card itself, check out our launch review of the Radeon HD 6870 and 6850.

AMD’s slightly better power consumption figures mean that buyers of the new green team can go green for the same green, while saving a little green over the life of the card (Ed.: Xzibit approves, dawg).

If you believe this makes the Radeon HD 6850 a perfect mid-priced performance part, you're in good company, as the number of folks willing to spend ~$200 is much greater than the market for flagship $500 boards.

If you're ready to buy, there are plenty of options available. Some vendors even offer multiple models. Choosing a card certainly doesn’t have to be difficult. We simply asked every major manufacturer to send its best, and those who didn’t have anything to offer beyond AMD's reference design chose not to participate. That left us with four souped-up models for your consideration.

Asus EAH6850 DirectCU Overclock Edition

Also known as model EAH6850 DC/2DIS/1GD5, Asus’ barely-overclocked Radeon HD 6850 includes enhanced cooling and advanced voltage control to let buyers decide the best performance-to-noise ratio.

A custom 9.7” circuit board faces its single six-pin PCIe power connector, easing installation into shorter cases. The DirectCU heat pipes add around 3/8” to the cooler’s height without affecting power connector placement, while a steel rail on the card’s top edge braces it against the extra weight of that cooler.

Placement into narrow cases or certain space-saving cube designs could potentially be problematic, but Asus’ design makes sense for the majority of performance PC builders.

Clocked only 15 MHz higher than AMD’s reference GPU specification and with no bump in memory frequency, the EAH6850 DC/2DIS/1GD5 offers builders a core voltage range of 0.95 to 1.35 V to do their own custom tuning.

Asus Overclocking Software

Asus SmartDoctor provides the EAH6850 DC/2DIS/1GD5 slide controls for voltage, GPU, and DRAM frequencies from its main menu. Users can select GPU frequencies between 600 and 1000 MHz, DRAM data rates between 3000 and 5000 MT/s, and core voltage from 0.95 to 1.35 V.

Advanced menus control overheat protection, alarm settings, and startup options.

A third advanced menu allows fan speed to be set at four thermal presets. Lower temperatures can help overclockers achieve higher settings.

SmartDoctor’s HyperDrive menu allows users to set an active GPU clock boost, activated when a 3D game is detected.

MSI R6850 PM2D1GD5

A quick look at the package reveals that our test unit is the OC Edition of MSI’s R6850 PM2D1GD5, separated from MSI’s reference-speed R6850 PM2D1GD5 only by clock rate.

The difficulties we had trying to set this article up at the end of October are bluntly revealed in this board. The OC Edition was initially listed on MSI’s Web site, only to disappear later. AMD’s Radeon HD 6850 graphics processor has little room to overclock at or near stock core voltage, and MSI cancelled its OC Edition card rather than risk instability.

Several factors caused us to keep MSI in the running for this roundup. First is that all our reference-clocked cards are dispersed across other test labs around the country, and including one as a reference point in this story is a useful way to show how much better overclocked samples perform.

Second is that the OC Edition card uses the same hardware as MSI’s standard-edition R6850 PM2D1GD5, so that underclocking allows the faster card to represent the performance level of its reference-clocked sibling.

The R6850 PM2D1GD5 uses the same 8.5”-long circuit board design as the HIS model from our launch article, with three of its four voltage regulators enabled. While the cooler extends the card’s overall length, the 8.6” distance from the card’s face to the end of its PCIe power connector will be more important to those who plan to use a small case.

MSI Overclocking Software

The current version of MSI Afterburner did not support voltage changes on this particular card as of this writing, though MSI has fixed this missing option in past cards through Afterburner updates. Other functions already work perfectly, including core clock, memory clock, fan speed, and advanced controls.

While manual changes stick between reboots, advanced fan maps require the software to be running. We believe MSI’s fan mapping is the best in the industry, and startup options allow it to launch automatically and be minimized to the task bar.

Monitoring, logging, and OSD options take up two more pages of the advanced menu.

MSI adds a screen capture function that automatically saves a game image.

Up to four overclocked configurations can be saved as profiles, with launch options that include game-based performance increases.

PowerColor PCS+ AX6850 1GBD5-PPDHG

One of the two highest-clocked samples in today’s roundup, PowerColor’s PCS+ AX6850 sweetens the deal with a free download of Call of Duty: Modern Warfare 2.

That “deal” is for a card that includes a four-phase voltage regulator to stably support its high clock rates on a circuit board that’s only 8.5” long. Identical in other ways to MSI’s PCB, the 8.6” distance between the slot plate and the end of its PCIe power connector allows reasonable cable clearance in tight cases. A plastic cover over the card's cooling fan extends its total mounting depth to 8.9".

Clock rates of 820 MHz GPU and GDDR5-4400 are sure to boost the PCS+ AX6850’s frame rates to performance levels not consistently supported by MSI’s similar three-phase design. A slightly larger heat sink keeps things cool under the added load.

PowerColor does not yet offer an overclocking utility for its Radeon HD 6850 cards, but users can always try one from a third party.

Registration gets PCS+ AX6850 buyers a month of VIP access to SuperStar Racing, an online multiplayer formula racing game.

Sapphire Toxic 6850

Sapphire does things its own way, beginning with an oversized blow-through cooler on its part number 100315TXSR Toxic 6850.

Sapphire buyers get a surprising array of upgrades, from the card’s dual mini-DisplayPort outputs (and single full-sized adapter) to its inclusion of a six-foot HDMI cable. Sapphire even throws in two four-pin Molex to six-pin PCIe power adapter cables.

The dual power adapters make sense in light of the fact that Sapphire puts two six-pin PCIe power connectors on its card. Sapphire’s entire design is borrowed from the Radeon HD 6870, stretching the card to a 9.6” mounted length, while adding the handy feature of a card that vents out the back of the case.

Even though Sapphire’s Toxic 6850 is oversized, it uses the same 820 MHz GPU and GDDR5-4400 clocks as the smaller PowerColor competitor in today’s comparison. Rather than include software, Sapphire adds onboard features and cables to make its Toxic 6850 unique.

Sapphire recently reintroduced its TriXX overclocking software to support this card, but the information didn't get to us in time for this review.  Users who prefer getting their tuning software directly from the manufacturer can download the utility here.

Test Settings

Gigabyte’s four-way CrossFire/SLI-supporting motherboard provides today’s test cards with all the performance they need, while opening up the possibility of future CrossFire scaling articles using some of today’s data.

Clock speed has often been our biggest bottleneck in achieving full graphics card performance. None of our games require more than three processing cores, yet Intel’s six-core Core i7-980X is already in the board and easily supports 4 GHz overclocked.

OCZ’s Z1000M provides over 88% efficiency across a very broad range of loads, at minimal noise.

This author’s argument against definitive noise measurements is that it’s almost impossible to remove background noise contamination from a running system. Yet, many readers insist that these measurements must be taken for relativity's sake. Using a “supersized” CPU air cooler that takes several minutes to warm up, the CPU fan has been temporarily disconnected during sound tests to get a somewhat-close approximation of each card’s true noise level.

Benchmark Results: Alien Vs. Predator

The AvP benchmark doesn’t show minimum FPS, but we didn’t see any frozen frames at 1680x1050 on any card. Sapphire edges out the same-clock PowerColor, but that’s likely due to slight variations within the benchmark itself.

Some users will need to disable anti-aliasing to play AvP at 1920x1080, our target resolution for cards that cost around $200.

2560x1600 tortures all three cards. The results likely rule out playing in Eyefinity mode across three 1280x1024 displays as well, since the pixel count is nearly matched.

Benchmark Results: Call Of Duty: Modern Warfare 2

Call of Duty runs at such high frame rates that other parts of the computer—such as CPU and memory clocks—often impede graphics performance. Overclocking helps, as it appears our system is at least fast enough to unleash the potential of a single Radeon HD 6850, even at our lowest test resolution.

All four cards breeze through even the highest resolutions in this test, making Sapphire's win superfluous.

Benchmark Results: Crysis

Crysis tortures every part of a modern system in spite of its age. Fortunately, 1680x1050 appears playable at Very High detail levels.

We had to take notes on minimum frame rates to assure playability with anti-aliasing enabled. MSI’s reference-clocked card drops to 18 FPS, while the rest barely meet our 20 FPS minimum qualification.

AA must be disabled to play smoothly at our 1920x1080 target resolution, where our notes say the slowest card stays above 20 FPS (minimum). The fastest card in the group, PowerColor’s PCS+ AX6850 only reaches 19 FPS with AA enabled.

Playing Crysis at 2560x1600 on any of these cards would require drastic drops in detail level.

Benchmark Results: DiRT 2

All four cards cruise through DiRT 2 at 1680x1050, with Sapphire’s Toxic 6850 taking the lead when AA is enabled.

We took notes on minimum frame rates and found the slowest card achieving 29 FPS with AA enabled, Ultra quality, and 2560x1600 resolution. Everyone passes, while Sapphire and PowerColor swap positions in performance leadership.

Benchmark Results: S.T.A.L.K.E.R.: Call Of Pripyat

The Call of Pripyat benchmark has, in the past, been an exceedingly strenuous workload for mid-range graphics cards. And yet, all four Radeon HD 6850s breeze through our 1680x1050 test.

Slow frame rates at our target 1920x1080 resolution with 4x AA enabled force us again to look at test notes. Sapphire and PowerColor each had a minimum 20 FPS, while Asus and MSI dropped to 19 and 18 FPS, respectively.

All four cards appear playable at 2560x1600, yet another look at our test notes showed a failing 14-15 FPS minimum for every card, with AA disabled. You're simply going to want something faster to play this game at these settings.

Benchmark Results: 3DMark Vantage

3DMark scores don’t mean much in the real world, but we did notice the scores in today’s Radeon HD 6850 roundup are very similar to those generated in our previous GeForce GTX 460 reviews. That similarity is partly due to disabling the PPU test, which we feel gives Nvidia an artificial and unrealistic advantage.

Power And Efficiency

Voltage and clock speed have adverse effects on power consumption, so we expect the fastest cards to fall to the bottom of our power chart.

Asus has surprisingly low power consumption compared to MSI, and even the slightest power advantage for Sapphire’s oversized card surprises us when compared to PowerColor. We wonder why Sapphire needed to include two PCIe connectors?

Sapphire edges out PowerColor in average frame rate, but margins this small could be pure variance.

This is where our match gets tedious. We wanted to base our efficiency number purely on averages, so we first averaged the No AA and 4x AA average frame rates of each card. We then calculated a class average by averaging our list of averages. For those following along, the class average is 49.8 FPS.

Dividing each card’s average by the class average puts its performance level on a percent scale, where “zero difference” is actually 100% and the top cards are higher. Following the same procedure for Full Load and Idle power gives us a similar percent scale where the most miserly cards have the lowest score. Dividing those results gives use a calculation of efficiency, where class average is 100%.

Yet no electronic component is 100% efficient. We subtracted 1 from each card’s result to move the chart scale by 100%, focusing only on the differences in efficiency. Asus’ lower-power “mid-speed” card is 3.9% more efficient than the class average, while Sapphire’s moderate power consumption makes it the most efficient “fast” card.

Heat And Noise

Heat and noise output are controlled not just by the GPU voltage and clock, but by the cooler design. Asus surprised us by having the lowest temperature, since the card was also relatively quiet. Sapphire takes second place when its fan is turned to max speed.

Quiet coolers are often sub-par when dealing with heat, but Asus proved the contrary with its DirectCU design. The chart below is not perfectly accurate, since some background noise was unavoidable.

Sapphire has the noisiest card. But looking beyond the card itself, there is an advantage to its noisier, blower-type fan. That advantage is case cooling, since the Sapphire Toxic 6850 is the only card in today’s comparison to vent most of its heat outside of the case. We’ve previously seen the consequences of using internally-vented cards in an extreme CPU configuration.

Following a similar method to that of our previous-page’s energy efficiency calculations, we compared temperature to noise for today’s cards. Note that the temperature division is “flipped over” in this case to give the lowest temperature the highest score, and that the “acoustic efficiency” chart does not account for performance differences.

Conclusion

Sapphire and PowerColor have the best-performing graphics cards, but who has the best performance-value? We averaged the price of all four cards, then divided the actual price by the class average to put each card on a percent scale.

Based solely on a comparison of price to performance, Asus wins. Yet, unlike PowerColor, Asus does not include a free game certificate with its card. If you don't already own Call of Duty and that bundled item is important to you, then the balance shifts toward PowerColor's offering. Likewise, Sapphire is the only company to include a free HDMI cable with its card. So if the card is going into a home theater environment and you're missing HDMI connectivity, Sapphire's inclusion is more meaningful. Both Sapphire and PowerColor provide marginally better performance than Asus, and the packaged bonus features for both Sapphire and PowerColor come at a mere $10 price premium compared to Asus.

So, which card would we chose? This tester would probably pick the Asus EAH6850 DirectCU Overclock Edition for its low noise. But then again this tester already has HDMI cables and more game licenses than he can ever use.

If this was a System Builder Marathon, I’d probably go with two Sapphire Toxic 6850 cards for their externally-vented design. Pushing graphics heat out of the case would allow a more aggressive CPU overclock, and the CPU is a big bottleneck in multiple-GPU configurations.

If I were building for someone else, I would probably choose the PowerColor PCS+ AX6850 for its high speed, moderately-low noise, and free game. That special someone would probably value the extra software worth anywhere from thirty to fifty dollars online. Now, where did I put that Christmas list?

Tuesday AMD senior marketing manager Peter Ross said that the company struck a deal with Valve Software to distribute the latest Catalyst drivers for the ATI Radeon graphics card on Steam. Although no specific details were provided, Steam users will (obviously) be alerted when the latest available drivers for their hardware becomes available on the platform, and will be able to install the package with just one click.

"Earlier this year at GDC we announced AMD Gaming Evolved, a program that demonstrates AMD’s ongoing commitment to PC gamers, PC game development and the overall PC game industry," Ross said in his official blog. "AMD and Valve teaming up to make sure that Steam gamers always have the most recent drivers for their graphics cards is just another way that we are showing our commitment to the promise we developed in our Gamers’ Manifesto."

According to Valve's most recent Steam Hardware Survey, 87-percent of DirectX 11 gamers consist of ATI Radeon or ATI Mobility Radeon hardware. For Nvidia fans, the same survey reveals that 59.11-percent of Steam gamers use Nvidia-based graphics while 32.98-percent use ATI-based GPUs. Only 5.6-percent of the systems surveyed actually support DirectX 11.

Given the popularity of Valve's content distribution network, it's surprising that AMD--or Nvidia for that matter--hasn't already entered into this type of distribution agreement. As Steam now serves as the premier content distribution platform for the PC, it seems only natural that both AMD and Nvidia should take advantage of its prominence.

Nevertheless, Ross said that driver distribution will officially begin on Steam with ATI Catalyst 10.9.

MSI plans to showcase its upcoming Graphics Upgrade Solution (GUS) at Computex next week, an externally-housed ATI Radeon HD 5670 1 GB card that connects to a notebook's ExpressCard slot. MSI's solution is one of many external options in the works--both Gigabyte and Shuttle recently displayed their offerings at CeBit back in March. As seen at the show, Gigabyte has chosen the docking station approach whereas Shuttle's I-Power GTX mini will use a special GTX connector.

As for MSI's solution, the company has tweaked the limitations of the ExpressCard connection by using a thicker, shielded copper cable and "improved" associated electronics. This supposedly allows for "over 70-percent" of the theoretical bandwidth as opposed to the 50-percent or less seen with other external solutions. MSI said that the next generation of GUS devices will take the simpler route by using USB 3.0, however the current model will still provide a nice performance boost over integrated graphics.

While this option may sound encouraging for gamers, there are a few positives and negatives associated with the technology. The external device will need a separate power source, and could prove cumbersome when traveling. However that's only a minor setback given that it can provide up to three additional video outputs. This means that consumers can set up a 4-display array using the output from the laptop (or use the laptop's LCD) and the three ports provided on ATI's card.

MSI said that the upcoming GUS will be offered in two forms: without the ATI card for just $99 to $109, or with the Radeon HD 5670 card for $169 to $229. As it stands, the device can't support cards more advanced than the Radeon HD 5670, as the GUS has a maximum power draw of 84 watts and does not have a 6-pin power adapter.

Professional graphics workers now have an entire new line of ATI FirePro graphics cards to choose from now. In addition to the FirePro V8800, which launched earlier this month at an 'affordable' $1,499, AMD has now expanded the family to include the ATI FirePro V7800, ATI FirePro V5800, ATI FirePro V4800, and ATI FirePro V3800 to fill different needs and budgets.

Like the rest of the current ATI GPU family, these FirePro cards support DirectX 11, OpenGL 4.0 and OpenCL along with ATI Eyefinity technology.

AMD also announced the ATI FirePro 2460 Multi-View, a low profile, quad display graphics solution designed for financial institutions. This graphic part sips power with an average board power consumption of 13W.

With Nvidia's new generation of GPUs based off of its GF100 architectures at our doorstep, the competition to be top dog in the graphics world is more heated than ever. Both top GPU makers now have DirectX 11 offerings for the next generation of games.

Now that Nvidia has just laid its GeForce GTX 480 and GTX 470 cards on the table, we wanted to know what AMD thought of the competition. This is what AMD had to say:

Four months after its launch, the ATI Radeon HD 5970 remains the undisputed performance leader. Six months after its launch, the ATI Radeon HD 5870 remains the clear winner at and below its price point. Add to that a top to bottom line-up of DirectX 11 graphics cards as well as cutting-edge features such as the immersive experience of ATI Eyefinity technology, and we are confident ATI Radeon graphics cards will continue to be the favorite choice of customers wanting the most advanced and efficient graphics products.-Dave Erskine, AMD spokesperson

Check out our full review of the Nvidia GeForce GTX 480 and GTX 470.

The Way It’s Meant To Be Played?

As hardware enthusiasts, it’s only natural to follow the news about upcoming launches. Power users begin formulating their opinions as soon as the first specs get tossed around, regardless of whether they’re official or not. Some of the longest forum threads I've ever read were rumor mill postings from very smart technologists trying to guess at the next generation's capabilities.

But in the case of Nvidia’s GeForce GTX 480 and 470, the company didn’t leave much to our imaginations. As far back as September of last year, the GPU giant was waxing poetically about its compute architecture. “Mmm, that sounds nice,” we thought. In January of 2010, it described in graphic detail how GF100 (the GPU centering on its Fermi design) would cut through DirectX 11 titles heavy on geometry. “Sexy! Can’t wait to try it out.”

So, you could say that Nvidia figuratively shot itself in the foot by talking too early, since it’s now late March—almost half a year after the Fermi disclosures were made—and the first desktop-class cards employing the design landed in our Southern California lab one week ago.

They’re not for sale yet. Nvidia says the initial round of cards, manufactured in-house, is shipping to the channel and should be available the week of April 12th (in two and a half weeks). Subsequent boards will come from the company’s partners sometime next month. What will availability look like? According to Nvidia, it’s shipping tens of thousands of GF100-based cards at launch, and by the middle of April, anyone willing to spend $500 on a GeForce GTX 480 or $350 on a GeForce GTX 470 should be able to buy one. Rest assured we’ll stay on top of pricing and availability in our Best Graphics Cards For The Money column.

Of course, AMD was similarly ambitious about availability of its Radeon HD 5800-series cards at launch, and we saw how that one played out. Even now, TSMC’s 40nm struggles spell availability issues on all DirectX 11-class cards.

History Is History

Regardless of the fact that Nvidia is entering the Windows 7/DirectX 11 market six months after its competitor, the company is here now. All of the delays have led up to this point, where we get two graphics cards to compete against the nine Radeon HD 5000-series models AMD has launched since its mad rush began. And, given the $500 and $350 price points, our comparison is further narrowed, since it’s only really the Radeon HD 5850, 5870, and 5970 that populate AMD’s enthusiast-class high-end segment. So, those are the cards you’ll see compared to these two in our benchmark suite. Two thirds of ATI’s DirectX 11 lineup is, as of now, still competition-less—providing DirectX 11 support is at the top of your buying criteria, of course.

We’re also adding AMD’s last-generation flagship, the Radeon HD 4870 X2, Nvidia’s, the GeForce GTX 295, and the company’s previously-quickest single-GPU card, the GeForce GTX 285.

Of course, the gaming community’s initial concern was that Nvidia conceptualized the Fermi architecture principally with compute performance in mind, leaving the GPU's traditional role of speeding up the latest games a secondary objective. There’s no question that the architecture (and by proxy, the GF100 GPU centering on it) is designed to drive the company’s Tesla family into increasingly-demanding supercomputing environments through the implementation of ECC memory and augmented double-precision performance. Nvidia will succeed here. The potential gains of parallelizing certain technical workloads are enormous, and Nvidia’s investment in software development means it has a formidable head start over AMD and Intel in this growing market. What remains to be seen is how the GF100 handles “everything else.”

Here’s where we answer whether the GF100 GPU can hold its own in gaming. The stage is set, the players are in their places, we have a handful of new tests to present, and the results are interesting, to say the least. Let’s meet Nvidia’s new GeForce GTX 480 and 470, and get this show on the road.

Nvidia’s GF100 Gets Scaled Back

We presented the GF100 GPU, based on Nvidia’s Fermi architecture, back in January. At the time, we forecasted the potential for a 2x performance boost transitioning from GT200-based cards like the GeForce GTX 285 to a GF100-based flagship. Of course, that was also based on the assumption we’d be seeing graphics cards equipped with the complete GF100. As it turns out, that isn’t the case.

In its full form, the three billion transistor GF100 features 512 CUDA cores (derived from four Graphics Processing Clusters [GPCs], each with four Streaming Multiprocessors [SMs], and each of those sporting 32 CUDA cores). But the GeForce GTX 480 employs 480 CUDA cores, while the GTX 470 is armed with 448—32-core drops in each case. Nvidia achieves this by disabling one of the GTX 480’s SMs and two of the GTX 470’s.

Because each SM also contains its own texture units and PolyMorph engine (the fixed-function logic responsible for the architecture’s exceptional geometry performance), both new cards sacrifice performance in those two areas, as well. The GeForce GTX 480 retains 60 texture units (down from 64) and 15 PolyMorph engines, while the GeForce GTX 470 offers 56 texture units and 14 PolyMorph engines.

GF100’s back-end is of course independent of the GPCs, so even with its scaled-back GeForce GTX 480 configuration, Nvidia is able to maintain all six ROP partitions. Each partition is capable of outputting eight 32-bit integer pixels at a time, totaling 48 pixels per clock. The GeForce GTX 470 isn’t as lucky; it loses one of the ROP partitions (dropping total pixels per clock to 40).

A complete GF100, with all of its ROP partitions intact, sports a 384-bit GDDR5 memory interface (one 64-bit interface per partition). The GeForce GTX 480 comes to the table with this exact configuration, serving up 256MB per interface for a total of 1.5GB of GDDR5 memory (that’s 177 GB/s, when you take the 924 MHz clock rate into account). Naturally, the GeForce GTX 470 gives some of that up. Its 320-bit interface plays host to 1.25GB of GDDR5 at a lower 837 MHz clock rate, which adds up to nearly 134 GB/s.

So there you have it. We’re looking at the same graphics processor presented back in January. As a result of yield issues, however, Nvidia’s new flagship and second-in-command are forced to employ a scaled-back version of the chip. While we’re certainly not expecting to multiply the performance of GeForce GTX 285 anymore, these should still compete aggressively with AMD’s Radeon HD 5870 and 5850 cards. Speaking of cards…

Meet The GeForce GTX 480 And 470

Although the GeForce GTX 480 and 470 center on the same graphics processor, as with AMD’s Radeon HD 5870 and 5850, they employ significantly different card designs.

GeForce GTX 480

The GeForce GTX 480 consists of a 10.5” PCB, making it half an inch shorter than the Radeon HD 5870. It requires one eight-pin and one six-pin connector to supplement the power drawn over PCI Express. Nvidia rates the board for a 250W TDP—significantly less than Radeon HD 5970, which itself barely ducks under the PCI-SIG’s 300W electromechanical ceiling—and recommends a 600W+ power supply.

But the story doesn’t end with board power. Though the GeForce GTX 480 would seem, on paper, less power hungry than AMD’s flagship, it’s clearly a challenge to keep cool. The card’s thermal solution is among the most aggressive I’ve seen from a reference design. A sink draws heat from the GPU surface and memory ICs. Five heatpipes conduct thermal energy away and into an array of aluminum fins (one is hidden in the picture), while a typical blower-type fan pushes air through the enclosed shroud and out the back of the card.

Most unique, perhaps, is that the surface of the card is actually part of the heatsink, above the fin array. Normally, this would be a part of the card you could grab onto when pulling it out of a system. But when I burnt my hand on it, I thought a temperature reading would be interesting. Turns out that, during normal game play (running Crysis, not something like FurMark), the exposed metal exceeds 71 degrees C (or about 160 degrees F). This will have some ramifications for running two cards in SLI, but we’ll get into that shortly.

GeForce GTX 470

The GeForce GTX 470 sports a more elegant package. It’s 9.5” long—again, half an inch shorter than the competition, AMD’s Radeon HD 5850. Two six-pin connectors provide all of the auxiliary power needed to supplement the PCI Express interface, maxing out at up to 215W (a little less than 30W more than a Radeon HD 5870). Gone are the heatpipes and exposed surface sink. Instead, the dual-slot card is completely enclosed, pulling air from the blower toward the rear and pushing it out the I/O bracket.

This board’s idle and load temperatures aren’t as aggressive as the GTX 480, though both cards are able to withstand GPU thermal thresholds as high as 105 degrees C. It’s actually interesting to watch these cards’ thermal properties in real-time. As load is applied, temperatures increase to the peak levels you’ll see at the end of this piece (97 and 96 degrees for the GTX 480 and 470, respectively), at which point the fan kicks up a notch to bring temps down by four or five degrees. Most other high-end cards we’ve seen get hotter and hotter, but eventually taper off just under the thermal threshold in response to faster fan speeds.

Tessellation And Anti-Aliasing

Two of the most notable areas Nvidia chose to zero in on with its Fermi architecture were geometry and anti-aliasing performance. Each is affected in a different way, but the concept is similar for both: enable more realism and improve image quality.

Tessellation: Standardized In DirectX 11

Tessellation is nothing new, but it’s suddenly the buzzword du jour now that it’s a feature of DirectX 11 (and not just limited to one vendor’s implementation). In the six months or so since Microsoft’s latest API made its debut, we’ve seen a couple of games with tessellation support, but they’ve frankly not lived up to what we we’d been amped up for by AMD. We’ve since seen some impressive demos from Nvidia, and the promise of what well-deployed tessellation can do certainly looks good.

Each of the GeForce GTX 480’s 15 Shader Multiprocessors and each of the GTX 470’s 14 SMs includes its own PolyMorph engine, which works with the rest of the SM to fetch vertices, tessellate, perform viewport transformation, attribute setup, and output to memory. From our earlier look at the GF100 GPU:

In between each stage, the SM handles vertex/hull shading and domain/geometry shading. From each PolyMorph engine, primitives are sent to the raster engine, each capable of eight pixels per clock (totaling 32 pixels per clock across the chip).

Now, why was it necessary to get more granular about the way geometry was being handled when a monolithic fixed-function front-end has worked so well in the past? After all, hasn’t ATI enabled tessellation units in something like six generations of its GPUs (as far back as TruForm in 2001)? Ah, yes. But how many games actually took advantage of tessellation between then and now? That’s the point.

Ever since the days of Nvidia’s GeForce 2 architecture, we’ve been hearing about programmable pixel and then vertex shading. Now we’re getting some very impressive shaders able to add tremendous detail to the latest DirectX 9 and 10 games (Nvidia claims a 150x increase in shading performance from the GeForce FX 5800-series to GT200). But I know we’ve all seen some of the terri-bad geometry that totally ruins the guise of realism in our favorite games. Purportedly, the next frontier in augmenting graphics realism involves cranking the dial on geometry.

…in order to facilitate the performance needed to make tessellation feasible, Nvidia had to shift away from that monolithic front-end and toward a more parallel design. Hence, the four raster and (15 or 14, depending on the card) PolyMorph engines. The company naturally has its own demos that show how much more efficient GF100 is versus the Cypress architecture, which employs the “bottlenecked” monolithic design—however, we’ll want to compare the performance of a title like Aliens Vs. Predator from Rebellion Developments with tessellation on and off to weigh a more balanced app. Up front, though, Nvidia claims that GF100 enables up to 8x better performance in geometry-bound environments than GT200.

It turns out the AvP isn’t the best game to measure tessellation performance—the technology is only applied to alien models, and it’s hardly a “geometry-bound environment.” Thus, we ran the Unigine Heaven 1.0 benchmark on all of our DirectX 11-capable cards to better gauge the efficiency of each architecture when tessellation is applied:

The way these cards fall is fairly expected, but it’s interesting to note the drop-off in performance when turning tessellation on. The Radeon HD 5970, 5870, and 5850 retain 66%, 65%, and 65% of their original performance, respectively. The GeForce GTX 480 and 470 retain 83% and 83% of their performance, respectively.

It’s fair to say that the Fermi architecture certainly gives up less performance than AMD’s design. But remember this is one component of next-generation games, and it’s not yet being used as intensively as the latest demos might suggest. For that, we’ll need mid-range cards enabling the technology in a broader market. AMD has them, but Nvidia still doesn’t. As DirectX 11 hardware sees greater adoption, game developers will dedicate more attention to enhancing their titles with it.

Update: If you want to get even more theoretical about the way GF100's distributed PolyMorph design handles tessellation scaling, check out the following chart:

This is the Unigine Heaven 2.0 benchmark cranked all the way up to Extreme. Though it's a fairly unrealistic implementation of tessellation (there's diminishing returns on the geometry created here), it conveys how well GF100 handles an increased tessellation load versus Cypress. GeForce GTX 480 is retaining 60% of its performance, while Radeon HD 5870 is down to 39% of its baseline frame rate.

Anti-Aliasing: Now with More Vitamin CSAA

The concept of burning through horsepower came up when we started talking about the Radeon HD 5800-series cards, and it’s relevant here, too. AMD and Nvidia both have all of this potential performance at their disposal. AMD is using features like Eyefinity to expand beyond 30” LCDs to render across larger surfaces, making the case for higher-end GPUs. Nvidia is using 3D Vision Surround to do the same thing, plus throwing more complex PhysX effects at the graphics processor.

In the same vein, GeForce GTX 480 and 470 also support 32x coverage sampling anti-aliasing, which decouples coverage from color/z/stencil data, and thus reduces the bandwidth and storage costs of applying AA versus MSAA. Using a little Battlefield: Bad Company 2 at 1680x1050, we tested Nvidia’s various anti-aliasing techniques.

Naturally, Nvidia’s GeForce GTX 285 is slower than the GTX 480, so look instead to the scaling chart, which shows the benefit of GF100’s enhanced ROP performance. As the AA is slowly cranked up, the GeForce GTX 480 loses less of its frame rate than GTX 285. And at the 32x CSAA setting, it’s impressive to see the GeForce GTX 480 running at nearly 80% of its baseline speed sans AA. Clearly, if you have the performance to spare, turning on anti-aliasing is a good way to improve image quality here.

Nvidia Surround, Display Output, And Video

For those who’ve come to know and love AMD’s Eyefinity technology, Nvidia Surround is a bittersweet effort.

On one hand, the launch of GeForce GTX 480 means that Nvidia, like AMD, is pushing enough graphics horsepower from a single-GPU card that gaming at 2560x1600 isn’t necessarily taxing. Thus, the company is enabling (via software) the ability to span games across three displays, akin to Eyefinity.

The good news here is that Nvidia Surround supports bezel correction so that display bezels function more like window frames and less like jarring interruptions between continuous objects on-screen. It also works on older GeForce GTX 200-series graphics cards—so anyone rocking an SLI-based configuration of, say, GTX 280s could conceivably take advantage.

The bad news is that GF100- and GT200-based graphics cards are limited to two independent display outputs each. Attaching three monitors requires two cards. Comparatively, AMD’s support for three (and soon to be six) LCDs from a single board is vastly superior. Granted, the number of people with three-monitor desktops is small, but I’m included in that minority, so it’s all the more relevant to me.

To Nvidia’s credit, it plans to go one step further by enabling stereoscopic mode across three displays—and it’ll be the only game in town for anyone willing to throw down for three new 120 Hz LCDs, two graphics cards, and a $200 GeForce 3D Vision kit. Expensive? Yes. Potentially very cool? Also yes. 3D Vision Surround support is not yet part of the GeForce 197 driver. Rather, it’s expected to emerge in the 256.xx release sometime next month, and we’ll need to follow up once the feature can be tested.

Display Output

As you likely surmised from the preceding discussion, GF100 proffers two display pipelines, which are integrated into the GPU (unlike last-generation’s cards, which employed the NVIO companion chip). Both the GeForce GTX 480 and GeForce GTX 470 include two dual-link DVI outputs and a mini-HDMI connector, two of which can be used at any given time.

This connectivity suite isn’t particularly unexpected given then prevalence of DVI and the two-output ceiling common to every card outside of the Radeon HD 5000-series. AMD’s emphasis on DisplayPort is driven predominantly by the way Eyefinity enables its three-plus outputs.

Video And GF100 In Your HTPC

With everyone focusing on compute and gaming performance, there’s been little attention paid to GF100’s A/V capabilities. Naturally, nobody is going to want a GeForce GTX 480 or 470 in a home theater PC. But what about the derivative cards sure to follow?

GF100 employs the VP4 engine introduced back in October of last year with the mainstream GeForce GT 220 cards. VP4 adds support for MPEG-4 ASP (Advanced Simple Profile) acceleration, in addition to the MPEG-2, VC-1, and H.264 codecs offloaded by VP3.  

Moreover, it’s no longer necessary to use an S/PDIF cable to enable HDMI audio input. GF100 (like the GT 220 and GT 240) supports HDMI audio over PCI Express, enabling multi-channel LPCM. TrueHD and DTS-HD Master Audio cannot be bitstreamed, as they can with AMD’s Radeon HD 5000-series and Intel’s Clarkdale-based Core i3/Core i5 CPUs. You can’t get lossless audio out of a GF100-based card, but again, that probably won’t matter in any way until the Fermi architecture is miniaturized in a big way.

Test Setup And Benchmarks

Benchmark Results: 3DMark Vantage

It might come as a bit of surprise to you that Vantage favors the Radeon HD 5870 over Nvidia’s GeForce GTX 480, and perhaps less of surprise that the dual-GPU Radeon HD 5970 and GeForce GTX 295 cards take first and second place. However, bear in mind that Vantage is a DirectX 10-class synthetic metric. While many of our readers like to see these numbers, they’re not necessarily representative of real-world performance.

Also bear in mind that we’ve disabled PhysX support here. Synthetic though this test might be, there’s clearly a huge weighting given to this capability, and we’re measuring graphics performance here.

Benchmark Results: Call of Duty: Modern Warfare 2 (DirectX 9)

We’ve long considered Call of Duty to be a processor-bound title, since its graphics aren’t terribly demanding (similar to Left 4 Dead in that way). However, with a Core i7-980X under the hood, there’s ample room for these cards to breathe a bit.

Nvidia’s GeForce GTX 480 takes an early lead, but drops a position with each successive resolution increase, eventually landing in third place at 2560x1600 behind ATI’s Radeon HD 5970 and its own GeForce GTX 295. Still, that’s an impressive showing in light of the previous metric that might have suggested otherwise. Right out of the gate, GTX 480 looks like more of a contender for AMD's Radeon HD 5970 than the single-GPU 5870.

Perhaps the most compelling performer is the GeForce GTX 470, though, which goes heads-up against the Radeon HD 5870, losing out only at 2560x1600 with and without anti-aliasing turned on.

And while you can’t buy them anymore, it’s interesting to note that anyone running a Radeon HD 4870 X2 is still in very solid shape; the card holds up incredibly well in Call of Duty, right up to 2560x1600.

Benchmark Results: DiRT 2 (DirectX 9)

This one almost had to be trashed altogether. Our DiRT 2 testing centers on the game demo, which includes a built-in benchmarking mode (the retail game does not come with this, and instead needs to be tested manually via FRAPS and a human pilot).

We noticed that the GF100-based cards were generating really high scores compared to the Radeon HD 5000-series cards, which seemed a little odd since both architectures are able to run the game in DirectX 11 mode. As it turns out, the GeForce GTX 480 and 470 aren’t properly detected as DX11 cards by the demo (though they function as expected in the retail game). So, we forced all of the boards into DX9 mode in order to get a look at performance in a DX9 title (rather than toss DiRT 2 as a benchmark altogether) .

Without anti-aliasing applied, the dual-GPU ATI cards take first-place finishes at 1920x1200 and 2560x1600. Nvidia’s GeForce GTX 480 performs well against ATI’s Radeon HD 5870, though its advantage slowly evaporates in moving from 16x10 to 19x12 to 25x16. The GeForce GTX 470 starts in front of the Radeon HD 5870, slips between ATI’s two fastest single-GPU cards at 1920x1200, and then falls behind the 5850 at 2560x1600.

Benchmark Results: Crysis (DirectX 10)

Crysis is perhaps the closest thing to a synthetic in our real-world suite. After all, it’s two and a half years old. Nevertheless, it’s still one of the most demanding titles we can bring to bear against a modern graphics subsystem. Optimized for DirectX 10, older cards like ATI’s Radeon HD 4870 X2 are still capable of putting up a fight in Crysis.

It should come as no shock that the Radeon HD 5970 clinches a first-place finish in all three resolutions. All three of the Radeon HD 5000-series boards we’re testing demonstrate modest performance hits with anti-aliasing applied, with the exception of the dual-GPU 5970 at 2560x1600, which falls off rapidly.

Nvidia’s new GeForce GTX 480 starts off strong, roughly matching the performance of the company’s GeForce GTX 295, but is slowly passed by the previous-gen flagship. Throughout testing, the GTX 480 does maintain better anti-aliased performance, though. Meanwhile, Nvidia’s GeForce GTX 470 is generally outperformed by the Radeon HD 5850, winning only at 2560x1600 with AA applied (though it’s an unplayable configuration, anyway)

Overall, the Radeons show the strongest in Crysis.

Benchmark Results: S.T.A.L.K.E.R.: Call Of Pripyat (DirectX 10)

In revamping the benchmark suite for this story, we added a number of DirectX 10 and DirectX 11 titles. Of course, testing DirectX 10-capable cards in a DirectX 11 game gives those boards a distinct advantage, since they’re forced down a previous-generation code path. So, while we have to address this in the next few tests, we ran the latest version of S.T.A.L.K.E.R. using DirectX 10, just to get an idea of how the latest cards compete against previous-generation hardware.

Similar to what we saw in Call of Duty, the GeForce GTX 480 begins by trailing the dual-GPU Radeon HD 5970 and GeForce GTX 295 boards, slightly outpacing ATI’s Radeon HD 5870 and the old 4870 X2. Shifting to 1920x1200, three dual-GPU boards beat the GTX 480. And by the time we’ve hit 2560x1600, the Radeon HD 5870 also turns in better numbers.

There’s a rub, though. While scaling with anti-aliasing mirrors the no-AA numbers at 1680x1050, ascending the resolution scale shows the dual-GPU cards incurring a bigger hit to the point where Nvidia’s new flagship actually takes second place at 1920x1200 (with a playable average frame rate, no less) and 2560x1600.

Similarly, the GeForce GTX 470 retains more of its performance at 1920x1200 and 2560x1600 than the Radeon HD 5870 or 5850, despite its placement on the chart.

Benchmark Results: Metro 2033 (DirectX 10/11)

Those are some of the ugliest charts I’ve ever posted, and the only explanation I can give is that this is a beast of a game with regard to its impact on performance.

You can switch from DirectX 10 to DirectX 11 from the display menu, which is nice. Perhaps the most glaring issue here is that the DirectX 10 cards get a massive performance boost versus the cards running in DirectX 11 mode, affecting the placement of the Radeon HD 4870 X2 and GeForce GTX 285/295.

Disregarding those scores, you have the Radeon HD 5970 on top at 1680x1050 and 1920x1200, followed by the GeForce GTX 480, Radeon HD 5870, GeForce GTX 470, and Radeon HD 5850. At 2560x1600, there’s only one ATI card able to run, and the GeForce GTX 480 is the only DirectX 11 board with enough memory to function with analytical anti-aliasing enabled. See the holes scattered throughout? Those are all settings where the given configuration would start, but then hover between zero and one frame per second. Clearly, ATI has some work to do enabling DirectX 11 with 4xMSAA turned on.

Just how significant is the performance fall-off from DirectX 10 to DirectX 11? Using the GeForce GTX 480, we can see the penalty is quite nearly double, regardless of the anti-aliasing mode you’re using. What could possibly exact such a poignant penalty? Turning on DirectX 11 enables a DirectCompute-based depth of field filter—a cool effect with significant performance ramifications. Clearly, you’ll need a graphics card capable of standing up to this title if you want to see it the way its Ukrainian developer intended.

It’s also worth noting that Metro 2033 supports PhysX. We didn’t benchmark with this feature enabled, but the effects affected by PhysX are numerous.

Benchmark Results: Battlefield: Bad Company 2 (DirectX 10/11)

Here again we have a title able to support DirectX 11, thus giving compared DirectX 10 cards a performance advantage—at least, you’d think so. But the GeForce GTX 285 and 295 both bring up the rear, leaving ATI’s Radeon HD 4870 X2 as the only contender to factor out of the apples-to-apples comparisons (and for what it’s worth, issues with the GeForce GTX 295 and Bad Company 2 seem to be fairly prevalent online).

All of our cards serve up what I’d consider playable performance at 1680x1050. Most interesting, perhaps, is that the GeForce GTX 480 and 470 sacrifice less of their performance in switching from 1x to 8x anti-aliasing, allowing the GeForce GTX 470 to jump in front of the Radeon HD 5870 with 8xAA enabled, even though the 5870 is faster without AA.

Shifting to 2560x1600 shows the Radeon HD 5970 still in its overwhelmingly-commanding lead. But while the GeForce GTX 480 would seem to fall to fourth place (behind the Radeon HD 5870), it actually scores a second-place finish once you turn on 8xAA.

ATI is able to maintain the fact that it sells the fastest card you can buy. However, with an expected price $200 below the Radeon HD 5970, the GeForce GTX 480 looks to be a contender here at $499. Of course, when you consider the Radeon HD 5870s already selling online at $419, Nvidia might have some work to do on its pricing.

Dual-Card Scaling: GeForce GTX 480 In SLI

We were able to get our hands on a second GeForce GTX 480 late in the game for some multi-GPU testing. And although we weren’t able to run the configuration through the complete game suite, we did run Metro 2033, S.T.A.L.K.E.R.: CoP, Crysis, Call of Duty: Modern Warfare 2, and DiRT 2 at 2560x1600 with 4x MSAA and Battlefield: Bad Company 2 at the same resolution with 8x MSAA.

Naturally, slinging a pair of high-end graphics cards makes the most sense to gamers running high-resolution screens. Almost across the board (Crysis is debatable), 2560x1600 with anti-aliasing enabled becomes a very real option. And once 3D Vision Surround becomes testable, you’ll almost certainly want the fastest cards you can get your hands on in order to achieve playable frame rates at, say, 5760x1200 in stereoscopic mode. Last-generation’s cards might be supported, but we're almost positive that they'll be too slow for the quality settings to which enthusiasts have grown accustomed, as scaling between the two cards isn’t expected to be as smooth as what we’ve seen from SLI.

The second chart maps out scaling from one card to two. In the worst-case scenario, SLI yields a 63% speed-up—not bad for a brand new game like Bad Company 2. In S.T.A.L.K.E.R.: CoP, the performance increase is more than double. Yes, we re-ran those numbers; they’re consistent. Despite the degree of horsepower wielded by the latest generation of GPUs, it’s still easy to slam into graphics bottlenecks. A pair of GeForce GTX 480s in SLI shows that at the highest resolutions, a second card can come close to doubling performance—the fastest frame rates you’re likely to see for a while, since power constraints mean it’s unlikely there is a dual-GPU card to anticipate.

Do bear in mind that there are some system builder considerations here. We had two expansion slots between cards on the Gigabyte X58A-UD5 motherboard used for testing. Nvidia recommends that, on boards with three slots available, you use the first and third, leaving room between. In short, GTX 480s back to back is a bad idea. A beefy power supply is a given, though Nvidia hasn’t yet recommended the best wattage for a dual-GPU setup.

Power Consumption And Heat

Here’s where things get dicey. I knew going into this story that the GeForce GTX 480 and 470 would be hot, power-hungry boards—Nvidia told me as much back in January. But measuring the extent of those values is an unscientific practice, at best.

FurMark is generally frowned upon as an unrealistic representation of peak power (a power virus, as AMD’s Dave Baumann puts it). However, it does serve as a theoretical worst-case scenario. Indeed, while the GeForce GTX 480 doesn’t use as much power as the dual-GPU GeForce GTX 295, it does out-consume the dual-GPU Radeon HD 5970 (no small feat, at 450W system power draw). Also, the GeForce GTX 470 uses significantly more power than the Radeon HD 5870.

Notably-missing from the chart is ATI’s Radeon HD 4870 X2, which we know from past reviews to chew up about as much power as Nvidia’s GeForce GTX 295. However, neither of the X2s in the lab seem to respond to FurMark at all anymore, running at a constant 13 frames per second or so and chewing up slightly-higher-than-idle power numbers. Maybe ATI “cured” that virus with forced lower frequencies in FurMark (the X2 was only able to hit 13 FPS or so, while our other cards were doing 40 or 50 FPS). But that doesn’t prevent the X2 from jumping into the 400+ watt system power range in actual games.

Heat jumps in FurMark as well, though it’s worth noting that none of these boards encountered heat-related stability issues. Keeping up with the thermals does mean the GeForce GTX 480’s fan ramps up fairly aggressively and does generate quite a bit of noise. However, we were unable to replicate that behavior in any real-world gaming load.

We’re actually a bit surprised about the idle power numbers as they were measured. AMD impressed us with the Radeon HD 5870’s 27W idle board rating, achieved in part by clocking its GPU down to 157 MHz and its GDDR5 memory to 300 MHz. Nvidia goes even further, dropping clocks to 50 MHz core, 67 MHz memory (270 MT/s data rate), and 100 MHz for the shaders. Nvidia doesn’t cite its idle board power, but an educated guess would still put the GeForce GTX 480 around 60W at those frequencies.

Just how do the GeForce GTX 480 and 470 size up in such a real-world load? Great question.

I ran all of the DirectX 11 cards in our story through the Unigine v2.0 benchmark, measuring average performance in frames per second. During the run, I had each configuration hooked up to a logger, polling power consumption every two seconds, yielding an average over the run. By dividing power use into average performance, we get an index that should give efficiency advocates something to think about.

Despite its aggressive power use, the Radeon HD 5970’s performance is enough to make it the most efficient board in the lineup, followed by the Radeon HD 5850 and Radeon HD 5870. Nvidia’s GeForce GTX 480 and GTX 470 pull up the rear. What I'm wondering is this: Nvidia rates the GTX 480 with a 250W maximum board power. AMD cites 294W for the 5970. Why do we keep seeing Nvidia's card using more system power?

Admittedly, these results are easily skewed—we can drive them one way or the other by hand-picking certain setting to cater to one or another architecture’s strengths. For example, turning off tessellation would give the Radeons a sizable advantage, since they take a more substantial hit when the feature is enabled. In fact, a couple of days before the launch, Unigine released v2.0 of the Heaven test, which adds even more emphasis on tessellation than the first revision used to generate our first batch of results. Thus, our numbers represent a best-case scenario for Nvidia; easing up on the tessellation load shifts the efficiency index even further in favor of the Radeon HD 5800-series cards, and we have charts demonstrating that, too.

Conclusion

As I was putting this story together, I asked senior editor Thomas Soderstrom what he wanted to know about GeForce GTX 480 and 470. He said, “price, performance, and power.” Reasonable enough. I’ll break it down similarly simply. There’s some good, some bad, and some ugly in play here.

First, the good—performance. Fortunately for Nvidia, it had a few targets in the Radeon HD 5970, 5870, and 5850 as it was generating specs. While we’re sure the company wishes it was shipping 512-shader cards instead of pared-down boards, it’s hitting high-enough clocks to make GeForce GTX 480 and GeForce GTX 470 generally-faster than Radeon HD 5870 and Radeon HD 5850. This is especially true when you turn on anti-aliasing, as the new GeForce cards take a much smaller hit than their competition. Nevertheless, AMD holds onto the single-card performance crown with Radeon HD 5970. On the other hand, Nvidia's position stands to improve moving forward, as its emphasis on DirectX 11 affects a growing number of titles.

What about the bad? Well, getting your feet in the door here costs $350. A flagship GeForce GTX 480 runs $500. Radeon HD 5850s recently dropped back down to $300 and Radeon HD 5870s can be found around $400. Though the GeForce cards are faster than their single-GPU competition, the premium is hard to swallow if power and display connectivity are important to you, and less-so if PhysX, CUDA, or 3D Vision are more interesting. Do we expect AMD to drop its prices in response? Don’t count on it. According to the company, demand continues to outstrip wafer supply, with the consequence of anticipated pricing stability. Should the GeForce GTX 480 and 470 maintain their MSRPs, you’re looking at a staggering of price and performance from the Radeon HD 5850 up to the 5970 (with Nvidia's GeForce cards in between).

Then there’s the ugly: power. Nvidia argues that the enthusiast space isn’t as sensitive to figures like power consumption, and that lofty load figures still only translate to a few dollars per year. However, when you have the system power of a single-GPU card outstripping the total power of a faster dual-GPU board (despite their respective max. board TDPs, which we really can’t vouch for), that’s something to think about. We’re not even talking about FurMark here—it’s the Unigine performance, power, and efficiency index that put things into perspective. Of course, that power invariably gets dissipated as heat, and thus the GTX 480, in particular, becomes a very hot card, cresting 160 degrees Fahrenheit on its surface during game play.

So, if you break everything down into power, price, and performance (meaning you’re most concerned with how GeForce GTX 480 and 470 stand up in today’s games), then today’s preview comes off somewhat underwhelming after the build-up.

Unfortunately, it’s difficult to account for the one thing Nvidia is perhaps best known for: its bead on the future and knack for shipping hardware chickens right as the software eggs are starting to hatch. Desktop graphics is but one of the company’s foci, alongside professional rendering, scientific computing, and increasingly, mobility. I’m not generally one to get super-excited about features promised to be commonplace tomorrow. However, Nvidia’s investment into software (from compiler support to math libraries to development toolkits) suggests a substantial advantage in maximizing its hardware in all three key segments. Love it or hate it, Nvidia’s developer relations team is the reason a game like Metro 2033 runs the way it’s supposed to at launch. That’s the intangible we can’t convey through benchmarks. Close cooperation with GPGPU vendors, getting into Adobe’s next-gen Mercury Playback Engine, zero-day support for more games, and so on.

We look forward to going into more depth on 3D Vision Surround and SLI—both of which are going to be attractive features for gamers with deep pockets. We also have a GPU/CPU compute story in the works that’ll pit cards from Nvidia and AMD against Intel’s 12-thread i7-980X. In the meantime, you’ll have to wait for GeForce GTX 480/470 availability anyway, since the cards won’t be available for another couple of weeks. Think about what performance means in relation to power consumption—that’ll likely be the underlying theme that becomes most important as derivative designs of GF100 are planned.

Additional Reading: Breaking Down GF100

A quick look at the diagram of the GF100 is enough to see extent of the changes in comparison to the GT200.

The diagram of the GT200 shows a set of functions shared by the entire GPU, such as the setup engine and the Streaming Processors Array. Conversely, on the GF100, four large Graphics Processing Clusters (GPCs) can be seen, and each GPC is largely independent.

Another remarkable point is the presence of a large L2 cache memory shared by the entire GPU. Now let’s look at the characteristics of the two new cards offered by Nvidia:

Taking into account the final clock frequency of the cores, the processing power works out to 1.344 TFLOPS with FMA (Fused Multiply-Add), compared to 2.72 TFLOPS for AMD's Cypress processor. However, you have to keep in mind that processing power is not everything when it comes to GPUs; it’s only one factor among many. GT200 was technically weaker than RV770 when it came to processing power, but it still dominated it in most games. That’s explained by the fact that the AMD's architecture is less flexible than Nvidia’s in the way it handles instructions in parallel. It uses VLIW units, which require a powerful compiler to extract the parallelism. The trade-off is that AMD can offer a high number of stream processors.

There’s another disappointment: the shader frequency. It has actually been decreased compared to GeForce GTX 285, down to 1,401 MHz from 1,476 MHz. On the other hand, GPU frequency increases from 648 MHz to 700 MHz.

We’ve been critical of AMD on several occasions for its modest texture unit and ROP enhancements, but we’ve never complained about Nvidia. With GF100, for the first time, a new Nvidia GPU actually has fewer texture units than its predecessor (60 compared to 80). But in practice, the units have been reworked, in particular, at the level of the memory subsystem. They're also clocked faster. Still, in certain specific tests, the difference is likely to have a noticeable impact. Let’s look at a simple quad texturing test, for example:

Here we see a difference of -11% between GeForce GTX 480 and GeForce GTX 280, which confirms our theory. On this very simple test, the Radeon HD 5870 is also 67% more powerful than the new Nvidia card. But we’ll get back to that point later.

Fortunately, there are no unpleasant surprises when it comes to the ROPs. Even though the use of a 384-bit bus (as opposed to a 512-bit interface on GT200) might suggest a reduction in the number of units, Nvidia more than compensates by doubling the number of ROPs per partition. The GT200 had eight partitions with four ROPs each, whereas GF100 now has six partitions of eight, for a total of 48 ROPs, enabling Nvidia to take the advantage back from AMD.

Additional Reading: SMs, Scheduler, And Texturing

Whereas AMD was able to limit itself to a few minor revisions with Cypress, Nvidia needed to radically change the G80 architecture it introduced over three years ago. The organization of the various units has been completely reworked. GT200 employed what Nvidia called TPCs (Texture Processor Clusters), consisting of one texture unit and three Streaming Multiprocessors (SMs). Nvidia maintains the term Streaming Multiprocessor with the GF100, but these units are now much more powerful.

Streaming Multiprocessors

With GT200, a Streaming Multiprocessor was made up of one 8-way SIMD units, two special function units, and one double-precision unit. With GF100, they now have two 16-way SIMD units and four special function units. So, there is no longer an ALU specifically dedicated to double-precision. FP64 calculation is now carried out by the same units at half the rate.

The increase in processing power is not the most notable change; the texture units are now directly implemented in the SM, whereas beforehand they were decoupled (three SMs shared eight texture units). On GF100, each SM has four texture units of its own, which also explains why their overall number has decreased compared to the preceding architecture (eight units per TPC on GT200 [or a total of 80], compared to four units per SM on GF100 [or a total of 60 for the GTX 480]). Another new feature is the 16 load/store units, enabling addresses to be calculated in cache memory or in RAM for 16 threads per cycle.

Concretely, this reorganization allows Nvidia to offer a much more elegant and more effective architecture than the previous one. The SMs are really independent processors, whereas beforehand they relied on a memory subsystem shared by groups of three SMs.

The size of the register file has also increased. Instead of 16,384 32-bit registers per multiprocessor, there are now 32,768. At the same time, the number of active threads per multiprocessor has increased compared to the GT200, from 1,024 threads (24 warps of 32 threads) to 1,536 (48 warps of 32 threads). Thus, the number of registers available per thread is 21 compared to 16 beforehand (and 10 on G80). Now let’s test the increase in processing power using a few shaders.

On a now-very-simple shader test, lighting per pixel in DirectX 9, the GeForce GTX 480 is 55% more efficient than the GeForce GTX 280. This increases to 82% with a heavier shader using procedural textures.

Dual-Scheduler

Since the appearance of its G80 architecture, Nvidia has been saying that its multiprocessors are capable of executing two instructions per cycle in certain circumstances: one MUL and one MAD. We weren’t able to confirm that with the early versions of the design, and even on the GT200 it was especially difficult to demonstrate. In practice, the famous dual-issue Nvidia was talking up wasn’t observable; the chip could execute only one MAD per cycle.

That’s not so important anymore because Nvidia’s GF100 has two schedulers that enable execution of two instructions per cycle without the constraints that limited previous architectures. As we saw earlier, the 32 stream processors in each multiprocessor are, in fact, arranged in two groups of 16 units, and each of these groups can execute one independent instruction. The recent history of CPUs has proven that superscalar execution is very complex to implement. But Nvidia has a major advantage in this area: GF100 doesn’t attempt to extract parallelism from a single instruction flow, with all the possibilities for error that implies. In fact, the multiprocessor selects two warps and launches execution of one instruction from each of them on the SIMD units. Since the warps are totally independent, they can be executed in parallel without any risk.

Most instructions, whether FP32 calculation instructions, integers, or load/store, can be executed simultaneously. The only exception to that rule is double-precision calculation instructions, which use all 32 stream processors and can’t be executed simultaneously with any other type of instruction.

Texture Units

Though the number of texture units has decreased, Nvidia has completely redesigned them in its current-generation architecture to improve performance. As we said, they are now built into the multiprocessors, which avoids having to share them among several multiprocessors, with all the loss of efficiency that involves. The L1 cache dedicated to texture units has also been redesigned, and though its size is unchanged from GT200 (12KB), Nvidia says that it’s much more efficient. Finally, the texture units are now clocked at the GPU frequency, whereas previously they operated at a lower frequency.

The result: on this test, which measures texture access performance (useful for displacement mapping, for example), the GeForce GTX 480 tested 75% more efficient than the GTX 280.

Obviously, the new units support the new BC6H and BC7H compression formats and the Gather instructions required by Direct3D 11. And performance with more contemporary pixel shaders has indeed increased, though that’s really a matter of catching up with the competition.

Additional Reading: Memory Hierarchy, Setup Engine, Tessellation

A Memory Hierarchy Worthy Of A CPU

For a long time, the memory subsystem was the GPU’s “red-headed stepchild.” Admittedly, memory access with early GPUs was so simple that a memory hierarchy wasn’t really needed. A simple texture cache and a small vertex cache (pre- and post-transformation) were all it took. With the arrival of GPGPU, things have become more complex, and a memory subsystem that basic isn't enough.

To meet these new requirements, Nvidia introduced Shared Memory with the G80. Shared Memory is a small, extremely-fast memory area built into each multiprocessor that enables data sharing among threads. Even though using this memory area was totally the responsibility of the programmer (and consequently required additional effort), the concept was well-received and reappeared in subsequent AMD GPUs. Finally, Microsoft integrated it into Direct3D 11, and Khronos into OpenCL.

But while this kind of memory, managed manually by the programmer, offers excellent performance with algorithms where memory access is highly predictable, it can’t supplant a cache when memory accesses are irregular. With GF100, Nvidia tried to combine the best of both worlds by offering a 64KB memory area per multiprocessor. This memory area can be configured in two different ways:

• 16KB of L1 cache and 48KB of Shared Memory
• 48KB of L1 cache and 16KB of Shared Memory

This L1 cache can be used as a buffer to back up the registers when space is short, which is not a bad thing. While their number has increased compared to GT200, the increase isn’t commensurate with the increase in the number of stream processors.

But that’s not all. As we saw earlier, Nvidia also adds a true L2 cache to the GPU. GT200 did have a 256KB L2, but it was accessible only for reading, and only by the texture units. With GF100, Nvidia gives its GPU a general-purpose L2 cache--that is, accessible in read/write mode--for texture instructions and for load/store instructions. This cache also replaces the ROP cache and the pre- and post-vertex transform cache of preceding GPUs. The L1 texture cache, due to its somewhat special functioning, is still present and is the same size as on GT200 (12KB).

Finally, A Parallelized Setup Engine!

The benchmark of one triangle generated per clock cycle has finally been surpassed! We’ve been waiting for it for a long time (since the arrival of unified architectures, in fact), representing a great leap forward in geometry power. For a short while, we even thought that RV870 had paved the way, but that was an unfortunate letdown. But now Nvidia is stepping up by multiplying the number of setup units, and it's being generous. There are not two, but four rasterizers!

Here’s where the new organization of the units shows its full potential. As we said, each GPC contains most of the functions of a GPU (except for the ROP units), and can consequently function independently. What that means is that each GPC has its own setup unit, which can rasterize one triangle per cycle (at 8 pixels per cycle). In practice, the peak performance is unchanged: 32 pixels are still generated per cycle, as with GT200 and RV870, but the effectiveness of the new architecture will be evident when numerous small triangles are being handled.

Like most tech companies, Nvidia has a marketing department that's always eager to come up with exotic names for familiar concepts, so it includes all of the operations performed on vertices within a GPC using the PolyMorph Engine moniker.

The geometric pipeline is as follows: the vertices are taken from a buffer, then passed to the multiprocessors, which perform the vertex shading. The result is then passed on to the next stage of the PolyMorph Engine (the Tesselator), which is not directly programmable, but whose operation can be controlled by the Hull Shader, upstream, and the Domain Shader, downstream on the multiprocessors.

Tessellation

A little review of tessellation is appropriate here. In Direct3D 11, Microsoft introduced three new stages to its graphics pipeline (in green in the illustration): the Hull Shader, the Tessellator, and the Domain Shader. Unlike the other stages in the pipeline, they don’t operate with triangles as primitives, but with patches. The Hull receives the control points of a patch as input, determines certain parameters of the Tessellator (for example, the TessFactor, which indicates the degree of tessellation). The Tessellator is a fixed-function unit, and consequently the programmer doesn’t control how the tessellation is performed; it sends the points it generates to the Domain Shader, which can apply operations to them.

Tessellation has multiple advantages. It enables very compact storage of geometric data. Only the control points of the mesh are stored in memory, and the additional vertices generated remain in the graphics pipeline, which saves memory and bandwidth. Another advantage is that tessellation enables the use of dynamic LOD (levels of detail). The closer a mesh is to the camera, the more detailed it needs to be. Conversely, if it’s farther away, using a huge number of triangles is not only superfluous, but it can also be a huge drag on performance since the rasterizers and shader units aren’t at ease with triangles that are too small.

Tessellation also has advantages in terms of animation; animation data is necessary only for the control points (the "coarse" mesh, in simple terms), whereas on a very complex mesh, it should be be available for each vertex. Consequently, the gains at this level can be used by adding more animation data per control point, improving the quality of the animation.

But let’s be clear: while Microsoft, AMD, and Nvidia try to make the situation sound idyllic, in practice, tessellation is not a magic wand that automatically increases a game's level of geometric detail. Implementing tessellation properly in an engine is still complex and requires work to avoid the appearance of visual artifacts, especially at the intersections between two surfaces with different tessellation factors.

For an analysis of tessellation performance, check out this page earlier in our review.

Additional Reading: Geometry, Raster, ROP, And GPGPU

Geometry Shader And Raster Engine

After our little sidebar on tessellation, let’s continue the tour along Nvidia's geometry pipeline. The last stage consists of the Geometry Shader, which first appeared with Direct3D 10, allowing vertices to be added to or taken away from primitives. We’re a far cry from T&L with our dear old GeForce (it’s been 10 years already).

This is an area where the new GeForce GTX 480 has evolved the most compared to the previous architecture, with no less than a 311% improvement in RightMark's Hyperlight.

Finally, the PolyMorph Engine performs the viewport transform and perspective correction calculations before passing the vertices and all of their attributes to the Raster Engine. The Raster Engine is made up of three main stages. First, the equations for the edges of the triangle are calculated, and the triangles that are not facing the camera are rejected. Then, the rasterizer generates the pixels (and the samples in the case of MSAA) covered by the triangle before passing all the data to a Z-Cull unit, now a familiar entity, which avoids performing pixel shading on hidden pixels through the use of a hierarchical Z-buffer.

Reworked ROP Units

As we saw earlier, Nvidia increased the number of ROPs, but it also made a few changes to them. The units' "graphics" performance hasn’t been changed (one 32-bit integer pixel per clock cycle, one FP16 pixel over two clocks, or an FP32 pixel over four clocks), but Nvidia has greatly optimized atomic operations (that is, memory operations carried out in one single transaction) with no interruption possible. This type of operation is extremely useful in parallel programming, when several threads can attempt to access the same resource. Nvidia claims very big gains--up to 20x in the case of atomic operations to a single address and 7.5x for contiguous memory regions, even if, in practice, these gains are probably more the result of using the L2 cache, rather than any substantial modification of the ROP units.

Nvidia also claims a substantial improvement to its compression algorithms, resulting in better efficiency with 8x anti-aliasing.

Finally, GF100 has a new 32x Coverage Sampling Anti-Aliasing (CSAA) mode and the possibility of combining CSAA and transparency multi-sampling to improve anti-aliasing of transparent surfaces. An interesting detail: until now, the number of pixels generated by the setup engine and the number of ROP units increased together. Now, GF100 has 48 ROP units and its rasterizer can generate only 32 pixels per cycle. That may seem like an odd choice at first, but in practice the latest games usually aren't played on high-end hardware without MSAA, and they often use floating-point frame buffers, thereby imposing a heavier workload on the ROP units, which take several cycles to process the pixels. So, the increase in the number of ROP units is justified, but in certain very simple rendering passes they’ll be underutilized by the rasterizers.

GPGPU

It’s hard to discuss the GF100 without talking about GPGPU, since Nvidia focused on it so much during its presentations of the new architecture. When the company designed G80, the GPGPU market was in its infancy. Nvidia’s choices proved to be good ones, both from the point of view of hardware (with Shared Memory) and software (Compute Shaders and OpenCL have programming paradigms that are very close to CUDA). But it’s impossible to offer a perfect solution on the first try, and Nvidia continues to develop CUDA, adding support for double-precision and atomic instructions with GT200. But those were just incremental improvements. With its Fermi architecture, Nvidia was able to take advantage of all the expertise gained from several years of work with CUDA to offer a much more powerful solution.

The first point that has been greatly improved with GF100, and that will directly benefit GPGPU applications, is support for double-precision arithmetic. As we saw earlier, the addition of double-precision support to the GT200 really looked like a quick and dirty solution, added just to stake out the territory. With a single 64-bit unit compared to eight 32-bit units, the GT200’s double-precision performance wasn’t really up to snuff. Without using dedicated units, AMD was even able to gain the advantage by lowering the performance of its GPU by "only" a factor of four with DP, which was only half the performance hit on Nvidia's architecture.

But with GF100, Nvidia completely reworked the architecture, and the dedicated MAD unit has been done away with. Now, the same units handle single- and double-precision calculation, and performance is reduced only by half with double-precision. So, the impact on performance is much more reasonable comparable to what’s observed on CPUs when using SSE. The advantage of using a GPU for this type of calculation should now be attractive enough to motivate programmers to do the necessary recoding.

While we’re on the subject of floating-point calculation, note that GF100 supports the most recent IEEE 754-2008 standard, with all of the required rounding algorithms and floating-point multiply-add double (FMAD) instructions. It maintains the precision of the calculation throughout and does only a single rounding, unlike the classic multiplication-addition (MAD) instruction, which performs two roundings. Note, however, that its direct competitor, AMD's RV870, is not really behind in this area because it also supports the latest floating-point standard and the FMAD instruction.

We described another advantage favoring Fermi in GPGPU environments on the previous page: a new memory hierarchy. In certain cases, a little scratchpad RAM can work miracles, but there are situations where nothing can replace cache memory. Also, GF100 has optional support for ECC memory, and all of the GPU’s internal memory (L1 and L2 cache, plus shared memory) is also protected.

Nvidia corrected two bottlenecks that could slow the performance of its chips in GPGPU mode. Like a CPU, a GPU gives the illusion of executing several tasks in parallel by alternating between them, with each task being given a portion of the GPU time. The main difference between CPUs and GPUs is that, on the latter, switching tasks is extremely expensive. Nvidia grappled with this issue in designing GF100, optimizing these operations. Context-switch time is now down to less than 25 microseconds. With that degree of improvement, frequent inter-kernel communication is now feasible where it wasn’t before.

Another big improvement: up until now, the GPU could only execute one kernel at a time on the entire GPU. With large kernels, that wasn’t a problem, and all resources were used. But with small kernels, it was possible for part of the GPU to go unused. GF100 is now capable of executing several kernels in parallel (in practice, up to one per multiprocessor), which results in more efficient use of the GPU, even with large kernels. That’s because when larger kernels reach the end of their execution, it’s possible for them to have an insufficient number of blocks to occupy the entire GPU.

Management of branching has been optimized too, with support for predication instructions. These execute two divergent portions of code in parallel before determining which one is to be kept. This avoids the additional cost of the branching instruction, which can be advantageous when the code to be executed is limited.

The final new feature is a unified memory space. Until now, PTX ISA 1.0 (the virtual instruction set to which CUDA programs are compiled) had three address spaces: the global, device- and system-wide space, the private local space for each thread, and the space shared by all the threads in a given block. The target of a load/store instruction had to be determined at compile time, which made it difficult to completely implement pointers whose target can change dynamically at run time. With the PTX ISA 2.0 supported by GF100, a single address space is used, which, among other things, makes support for C++ programs possible. C++ objects largely depend on the use of pointers to implement virtual functions, whose behavior can change at run time depending on the dynamic type of the object.

Let’s give credit where credit is due. After several months of stagnation, the little world of 3D has finally awakened, and we now have two Direct3D 11 architectures, designed with quite different approaches.

They're both a little on the large side for netbooks (12.1-inch displays), but they're neat little notebooks that boast AMD's Athlon Neo MV-40 and Athlon X2 L335 respectively. The U230-033 packs a 250GB HDD while the 040 bumps storage up to 350GB.

That's where the differences end, folks. They both feature ATI's Radeon HD3200 graphics, 2GB of RAM, a 6-cell battery, 802.11b/g/n WiFi, 3 x USB, VGA and HDMI out, a 4-in-1 card reader and a 1.3 megapixel webcam. Each weighs 3.3 pounds and runs Windows 7 Home Premium.

The U230-033 has a price tag of $429.99 and the U230-040 costs $479.99. Both are available today on Newegg.

Nvidia's GF100/Fermi GPUs, the GTX 470 and GTX 480, will be officially launched later this month on March 26. Until then, we're being teased by unofficial board shots and unverified benchmarks.

Today, however, we bring to you something completely official: a benchmark demo straight from Nvidia that compares the upcoming GeForce GTX 480 and the ATI Radeon 5870.

While Nvidia doesn't put the two video cards up against our usual battery of tests, it does chart the two in an demo that leans on hardware tessellation. With tessellation demands piled on, the GeForce GTX 480 performs significantly better. Other times, however, it seems that the Radeon 5870's performance will be on par with Nvidia's upcoming high-end GPU.

Until developers take advantage of hardware accelerated tessellation however, the 5870 still gives a GTX 480 incredible competition for being over six months old.

See the video below for more.

En Route To The Holodeck

Major changes to the PC graphics experience come few and far in between. Perhaps the last significant update was the move to multi-GPU rendering, a la ATI CrossFire. Of course, despite the well-known performance benefits of employing multiple GPUs, the technology has some obvious drawbacks. More than one motherboard slot is usually required for multi-GPU configurations, which in turn means that you wind up consuming two or more cards worth of power, rather than just one.

On the application front, you're free to span a desktop across multiple screens, running multiple apps across one or more displays concurrently. However, traditionally, spanning an application across more than one screen meant that it had to be non-accelerated. You could have size or you could have speed, but you couldn’t have both.

Times have changed. AMD’s new Eyefinity technology takes us to the next level in mainstream multi-monitor output. Known as “SunSpot” to only a handful within AMD up until its launch day, Eyefinity enables up to six monitors to be controlled from one card, enabling a massive “surface” area exceeding 24 megapixels. If you read AMD’s literature on Eyefinity, it says that “we are inexorably on the road to the ‘holodeck’ (as conceptualized on Star Trek).” Given that the Star Trek holodeck involved haptics (tactile feedback) based on force fields and such, this might be a tad hyperbolic. A better, if less-known, analogy might be the CAVE (Cave Automatic Virtual Environment), a small chamber comprised of three to six screens showing rear-mounted 3D projections. Dozens of universities use CAVE systems for research around the world, so imagining a consumer version of CAVE enabled by AMD’s technology isn’t at all far-fetched. If you’ve read Fahrenheit 451 and recall the three-walled, immersive home entertainment systems Bradbury imagined, that’s where Eyefinity (with a little help from ultra-high-speed broadband) is headed.              

If your primary PC is a single-screen notebook, the thought of having six monitors may seem overwhelming and excessive. So let’s be clear at the outset: Eyefinity is a means, not an end. If CrossFire was ATI/AMD’s method for helping consumers go multi-GPU, then Eyefinity is the next evolution of that. The object is not to pile on as many screens as possible. Rather, the object is to make computing “surfaces,” the real estate on which you visualize a computer-generated environment, as scalable as possible.

Today, DisplayPort is a key piece in making that scalability happen, as is a redesign in how multi-monitor computing happens on the video card. Let’s dig in and see what Eyefinity really is and does.

Leaping Beyond Two Screens

There are different ways to create a multi-monitor surface. Nvidia drew some eyeballs at CES 2010 with the demonstration of its 3D Vision Surround technology, but this approach required two boards, controlled and synchronized with software. Each card could still only output to two displays, as they’ve done for years. There was no increase in hardware-based surface scalability. The implementation still required two cards, an SLI-capable motherboard to run them, and more overall power compared to a single-card solution.

With Eyefinity, one card can drive up to six displays (depending on the model, of course). AMD states that all 5000-series GPUs support Eyefinity. However, it falls to the graphics card manufacturer to decide whether and how Eyefinity will be implemented on a given card. As of this writing, only the 5800 models have Eyefinity enabled in CrossFire mode, and it seems reasonable to expect that bandwidth plays a part in this decision. With the Radeon HD 5970 freshly placing AMD back in the lead for desktop GPU performance, the last thing AMD wants to risk is a bunch of 5400-series buyers trying to span Left 4 Dead 2 across three screens at maximum settings and wondering what why it isn’t playable. Having the display connectivity for multiple displays is one thing, but putting the muscle behind those pipelines is something else, and vendors need to pair their output paths and GPUs appropriately.

Just to be clear, the Radeon HD 4000-series GPUs and all of their predecessors do not support Eyefinity. The GPUs, good as they were, don’t have the horsepower or display output connectivity needed to fuel more than two high-resolution displays.

“For 2D applications, like for productivity work, the bandwidth load is no problem at all,” says Shane Parfitt, product manager in AMD’s GPU division. “Where you get into performance situations is 3D games with 3x the number of pixels in a 3x1 configuration. That’s definitely harder on the GPU. Our GPUs right now are the most powerful on the planet. We test with 30" panels—the 2560x1600 panels with over 6 million pixels each. Sometimes, you might have to turn down the anti-aliasing settings in order for a game to run at at least 30 fps. An average case of 19x12 running across three panels, we find, performs pretty well, especially with the higher-end cards.”

What about harnessing two or more GPUs to buy more multi-display bandwidth?

“We’re just now rolling CrossFire support out,” says Parfitt. “We had a lot of demand from the market for this feature, so we rolled it out in a hotfix driver. But it became an official feature in the February Catalyst release.”

Specifically, in November, the Radeon HD 5970 became the first GPU to support Eyefinity with CrossFire at a time when only 20 or so applications had been validated for CF/Eyefinity compatibility. Only three months later, the feature was enabled across the board. It deserves mention that making this happened required massive testing and troubleshooting effort on AMD’s part, and its driver crew should get bonus kudos for their beyond the call of duty labors.

Counting Pipes

Let’s say we’re dealing with an adapter able to keep three display outputs well-fed, such as a Radeon HD 5870. In this configuration grid, you can see the connectivity options for a four-port Eyefinity card. The card integrates two DVI ports, one HDMI, and one DisplayPort. Despite having four physical ports, the Eyefinity implementation allows for three actual output streams. One port is always disabled.

Under the hood, we can see the reason why only three displays are supported. The 5000-series designs host two internal clocks and one external clock. Legacy displays—now defined by AMD as VGA, DVI, and HDMI—each require their own internal clock. The magic of DisplayPort in this context is that it allows users to drive multiple displays from only one external clock. This external clock is what allows Eyefinity to fuel up to six monitors from a single card. If the card design accommodates it, all six monitors could be based from the same clock. However, given the profusion of legacy displays today and the relative dearth of DisplayPort screens, Eyefinity implementations currently provide for two legacy displays.

From this diagram of the display engine behind Eyefinity, you can see how a four-port card could be alternatively configured. You could have one VGA port, one DVI, and two DisplayPorts, for example. With the proper adapters, your configuration options expand considerably. Converting from legacy to legacy, as with the DVI to VGA example in configuration #5 above, requires a passive adapter. Similarly, according to AMD, “if only one or two DVI/HDMI monitors are attached through a DisplayPort connector, then only simple passive adapters are required.” A passive adapter performs no signal conversion; it merely acts as a pass-through. However, going from DisplayPort to legacy Dual-Link DVI or VGA requires an active adapter.

Screen Groups

What sort of display arrangements can you make with Eyefinity? You can use the Duplicated (cloned) mode to copy one desktop onto multiple screens. Laptop users do this all of the time for presentations. Alternatively, Extended mode lets you span one workspace across multiple screens, a grouping AMD calls a “single large surface.”

At its simplest, Eyefinity lets you do basic dual-screen setups, just as you would with a traditional dual-output card. Screens can be set in 2x1 landscape, 2x1 portrait, or 1x2 landscape. Interestingly, AMD kept its eye on streamlining the user experience without limiting reasonable options. For instance, you won’t find a 1x2 portrait group configuration because there’s no common use scenario that supports such a tall, narrow display. Why clutter up the configuration UI with an option for it? The same three orientations apply for triple-screen viewing: 3x1 landscape, 3x1 portrait, and 1x3 landscape.

Stepping up to four monitors, there are only two configurations offered: 4x1 landscape and 2x2 landscape. When you see shots such as the 3x1 portrait Battle Forge example, you have to wonder, “OK, why not 4x1 or 5x1 portrait?” Apparently, AMD’s research led them to decide that the number of people who might use such a mode didn’t justify including the option in the driver. And curiously, there actually are no five-screen modes.

For six monitors, there is only the 3x2 landscape option.

Now, what we’ve described here are your options for “display groups.” You can also create “Extended” groups. These are essentially separate desktop spaces that can run alongside your primary display group. You can see some examples in the image below. You might have a 3x1 portrait display group of 24" displays on your desk but have a separate 19" extended screen mounted to your wall for messaging apps and perhaps another 15" extended screen (making three groups total) for showing system temps, fan speeds, overclocking tools, and so on.

The theoretical maximum for today’s Eyefinity technology specs a maximum 8192x8192 (67.1 megapixels) resolution, figuring six 2560x1600 panels.

Software Configuration

Following ATI tradition, AMD uses the Catalyst Control Center (CCC) rather than a separate utility for configuring and managing all Eyefinity operations. Let’s say that CCC auto-detects that you have three monitors attached. These will show up in a row along the bottom area of the UI as disabled displays until you enable them. When you right-click a disabled display, you’ll have options to use that screen as a clone, make it the new active display in place of another, or extend a desktop onto it. Alternatively, pulling up the properties for a desktop group lets you select the total desktop resolution, color bit depth, rotation, and the refresh rate. The possible resolutions shown as options are based on the resolution modes for your monitors as detected by the driver.

One of your first tasks will be to create a display group by going to the CCC’s menu bar, clicking Graphics, choosing Desktops & Displays, right-clicking on the main display shown and picking Display Group > Create Group. If you want a group in portrait orientation, you’ll have to begin by rotating your initial display, then creating the group.

When you have three or more displays in a group, use the Select Layout option to pick a display group configuration. AMD’s drivers are smart enough to only display configuration options that are possible for your number of attached monitors. The CCC has a nifty shortcut in that if you pick a group configuration that matches the number of available displays, the drivers will automatically select all of the screens and lump them into a group for you.

In AMD’s marketing, you’ll find that all of the monitors shown in a display group are the same model. In the real world, this isn’t likely to be the case. You’ll probably have a mix of new and old displays you want to leash into a group and/or extended groups. You can use monitors with different resolutions, but Eyefinity will force all monitors in a display group into the lowest common resolution and orientation. Extended groups can be different resolutions, but obviously your main display group is the chief concern. Because you’re having to work with the lowest common denominator between the screens, this is why it’s always best to have your main display group comprised of identical screens.

If you’ve ever set up a multi-monitor config in the past, you know what a headache it can be to figure out which screen plugs into which port that’s seen as a given screen number by the driver, all of which you have to track so you can play musical monitors in the UI until the screens flow in the order you want. Eyefinity remedies that mess with a new wizard that runs for each new display group. In the UI’s main area, you’ll see a grid representing your monitor configuration. Your displays will black out, then the wizard will turn one blue. Just click on the grid cell for the highlighted monitor. The wizard does the rest, and you have a fully configured group in just seconds.

On the off chance that you need to regularly change your multi-monitor arrangement, be sure to create and save a profile for each configuration. You can even create a hotkey macro under the Activation tab to toggle between profiles.

Why You Want DisplayPort

Dual-Link DVI has been the premium multi-monitor interface of choice, but that’s likely to change now. Being digital, DVI doesn’t have VGA’s requirement for one digital-to-analog converter (DAC) per output port, yet it does require a dedicated clock source for each output. This design element is also true of HDMI.) According to AMD, the signaling demands of DVI require so many I/O pins from the GPU that extending beyond two displays was impractical. Engineers at ATI recognized this even back in 2004 and started working on ideas to move beyond DVI’s limitations—ideas that AMD inherited and continued to develop after the ATI acquisition.

The first version of DisplayPort was finalized in 2006; the current version 1.2 was only recently approved last December. The key advantage of DisplayPort is that there’s no need for a clock signal. Each data link is set to either 1.62 or 2.7 Gb/s per lane regardless of the timing used by the target display. DisplayPort simply buries the clock signal within the data stream. Freed from the need for dedicated, independent clock signals, a single clock source can regulate as many output streams as the GPU can support. DisplayPort uses the same number of pins as Single-Link DVI, but it can support full resolution on a 2-meter cable, 1920x1080p over a 15-meter cable, up to 17.2 Gbs per channel for higher resolutions, up to 16 bits per color channel, and multiple video streams over a single connection. So, if output to six monitors from one card sounds impressive, you have to wonder if future GPUs will leverage this multi-stream capabilities to multiply monitor counts even further. Consider the implications for being able to feed multiple monitors from a single DP port on a mini-ITX board or a notebook. As of version 1.2, you can even daisy chain DisplayPort monitors, feeding multiple monitors with only one DisplayPort connector. 
    

In our follow-up Eyefinity coverage, we’ll take a closer look at DisplayPort adapters. For now, know that adapters will allow streams from a DP port to display on HDMI or DVI devices. Like HDMI, DisplayPort can carry audio alongside video on the same connection. The DisplayPort++ feature allows for pass-through of a combined audio/video HDMI stream, but DP++ is not universally supported, so if this is a feature you suspect you’ll need, ask the adapter manufacturer before buying. Note that adapters only work from DisplayPort to legacy formats. You can’t output VGA, DVI, or HDMI on a DisplayPort monitor.

Predictably, DisplayPort is fully compliant with prior DRM technologies and some forward-looking DRM besides, so don’t look for DP to make ripping any easier.

On the monitor side, Dell was the first and still most ardent supporter of DisplayPort. In 2008, Apple jumped all over Mini DisplayPort, which it designed. Mini DisplayPort became part of the DisplayPort 1.2 spec and can support resolutions up to 2560x1600.

With AMD’s Radeon HD 5700- and 5800-series cards, DisplayPort is supported throughout the entire product line, and a host of major notebook OEMs now or will soon support the format. Clearly, DisplayPort is on its way to being the next major display standard throughout the computing world. Further investment in legacy formats looks more risky with each passing week.

Productivity

You would think that with the plummeting price of LCD screens, the need of the PC industry to sell more hardware, and an overarching desire for increasing productivity in the workplace, there would have been countless studies about the impact of multiple monitor use in the workplace. The bizarre reality is that you could probably count the number of such studies on one hand. The seminal piece of research was a cooperative effort in 2003 between (what was then) NEC-Mitsubishi, ATI Technologies, and the University of Utah. In assessing single- versus dual-monitor (18" LCDs) usage in everyday work environment applications, “participants in the study considered multi-screen configurations significantly more useful than single screens and preferred multiple monitor setups on every measure of usability. They found them 29 percent more effective for tasks, 24 percent more comfortable to use in tasks and found it 39 percent easier to move around sources of information....Overall respondents in the study were 10 percent more productive using multiple monitor set ups.”

Also in 2003, Microsoft research designed a 42" wide, 3072x768 display called Dsharp. The device utilized three DLP projectors casting onto curved Plexiglas. “The first study revealed that the users' productivity increased by 9 percent,” notes Microsoft’s press release. “Further studies showed even greater increases—at times up to 50 percent for tasks such as cutting and pasting.”

Then came an inexplicable five-year research hiatus. Finally, Fujitsu Siemens teamed up with the Fraunhofer Institute for Industrial Engineering and in early 2009 released a study detailing how 67 people were divided into three groups to complete a given set of productivity tasks. Group 1 used a single 19" display. Group 2 used a 22" widescreen. Group 3 used a triple-monitor (19") array. Because of learning, Group 1 improved by 1.9 percent. Group 2 took an 8.4% jump, and Group 3 blew the doors off with a 35.5% gain.

Clearly, more desktop surface yields greater productivity, and why every monitor and graphics card company hasn’t drilled this into the public’s collective head over the last seven years or so is anyone’s guess. Fortunately, AMD has revived the message with Eyefinity. How is one Eyefinity-enabled 5000-series card at $100 going to stack up against a non-Eyefinity card pushing three or four monitors with ATI SurroundView (the ATI discrete-plus-graphics chipset hybrid approach)? Good question. We haven’t been able to run those tests yet, but our guess is that the significantly faster architecture of the 5000-series GPUs combined with their much lower idle power consumption will make for at least a few compelling reasons to prefer Eyefinity over ATI SurroundView.

Gaming

Tom’s Hardware readers should need no introduction to the idea of multi-monitor gaming. Whether you’re a racer, general, assassin, or anything else: the more you can see, the better you’ll play.

“You’re seeing more of the game,” says AMD’s Parfitt. “It renders to a wider field of view. Like in driving games, the center monitor is what you normally see, but with Eyefinity, the side monitors become like your side windows. In flying games, you don’t get as disoriented because you can see more of the ground. In strategy games, you can see more ground to see your enemy and what’s going on in other parts of the map. In FPSes, like Left 4 Dead, you now have peripheral vision. In almost every kind of game, it’s a new experience.”

We tend to be partial to FPS titles, and one of the inherent downsides in any shooter is the never-ending need to keep turning left and right to check what should otherwise be your peripheral vision. But with one monitor you’re essentially stuck with wearing blinders. Enemy AI can see you perfectly, but unless you’re looking right at your foe, you really don’t know what you’re dealing with. The ultra-wide field of view made possible by Eyefinity restores much of that lost peripheral vision.

“In Left 4 Dead 2,” says Parfitt, “it’s very co-op, so if you leave your pack of four, you don’t have a chance because zombies will swarm you from everywhere. You have to know where everybody is in order to survive. Having that peripheral vision is a huge advantage. In Call of Duty, if someone is trying to sneak up on you from the side, it’s easy to just turn and melee attack him. It’s almost an unfair advantage in multi-player.”

Admittedly, not every game is designed to handle Eyefinity at its finest. Games from smaller developers will sometimes stretch to fill all of that extra resolution, so content side monitors can look skewed and distorted. But games have to deal with small and jumbo panels alike, so developers have increasingly shifted to computing a complete 3D world and then render the content for however much of that world can be displayed in the established resolution. This has become the norm in major titles, which is why most games now look optimized for Eyefinity right off the bat.

Entertainment And Beyond

The home theater PC (HTPC) market is still a fledgling field, but you don’t need a Ph.D. to recognize the growing threat that Hulu, Netflix, and others represent to the established cable and satellite markets. Why pay $150 per month for what can be subscribed to online for a fraction of the cost? All that’s needed is a way to push that online content off the desktop and through the home theater.

Eyefinity can help. Until now, we’ve usually thought about HTPCs as a single display affair—one compact PC outputting to the TV. Higher-end configs might build a tiny, secondary LCD into the system chassis. But what if you weren’t limited to one big display? What if you didn’t have to constantly switch channels or streams, or you could be free of annoying picture-in-picture limitations? If you had an HTPC that could support three dual-tuner cards, why not run five sports games on five screens and then have ESPN.com running on a sixth screen—all from one graphics card? Hitherto, this would have required multiple HTPCs, but now a “man cave” can run from a single microATX form factor machine.

Admittedly, you’re sucking down a lot of power for all of those displays, but at least you can drop your system count and save several hundred watts in the process. Actually, the power story is even better because of power savings in the new 5000-series GPUs. For example, idle power consumption on the Radeon HD 4890 was 90W. With the Radeon HD 5870, idle power plummets to only 27W.

Today, many tuner applications don’t support multiple output instances, but perhaps that’s because there hasn’t been a mainstream platform capable of providing an affordable many-stream solution. Eyefinity opens up a lot of future possibilities. Modern Radeon GPUs can support multiple H.264 decode streams at once. If you don’t care so much about having a TV tuner blitzkrieg but want more content off the Web, Blu-ray drives, and/or your hard drive, you can now go multi-stream output without sacrificing resolution because everything has to display on a single screen.

Is Eyefinity a game-changer in the display space? At the high-end, probably so. The technology’s potential in gaming and multimedia have probably only just started to be imagined. And even in mainstream settings, particularly in businesses, what employer wouldn’t see the wisdom of even a 15% productivity increase in workers who go to triple-head display? The cost of such an upgrade used to be prohibitive, but if one controller can power 3D apps across every display at full speed for under $100, and if three 19" or even 22" LCDs can be had for less than the price of one 30" screen, the return on investment seems obvious.

In our Eyefinity follow-up, coming shortly, we’ll take a look at the developer efforts surrounding AMD’s new display platform and some of the finer details you’ll need to know when you’re ready to take the Eyefinity plunge.

AMD keeps on rolling out with the DirectX 11 GPUs from ATI. The company today introduced the ATI Radeon HD 5830 graphics card, which at $239 USD, fills the gap between the 5770 and 5850.

"The ATI Radeon HD 5830 graphics card makes enthusiast-level performance even more accessible to gamers, adding another compelling choice to the award-winning ATI Radeon HD 5800 series," said Matt Skynner, vice president and general manager, AMD Graphics Division. "Cutting-edge features such as full DirectX 11 support, ATI Eyefinity multi-display capabilities and ATI Stream technology position the ATI Radeon HD 5830 graphics card to become a favorite with the gaming community."

We've put the Radeon HD 5830 through our own range of tests and found it to be a worthy offering, especially for those who were unable to find a Radeon HD 4890.

Incredibly similar gaming performance makes comparisons between the Radeon HD 5830 and Radeon HD 4890 easy. The similarities don't end there, as both cards were launched at similar price points. Almost certainly, given enough time and (hopefully) some competition, the Radeon HD 5830 will drop in price and deliver the same price/performance ratio that the Radeon HD 4890 has spoiled us with. And let us not forget that the 5830 delivers more than game performance alone. It uses half the idle power of the Radeon HD 4890, notably less power under load, and supplies us with all of the Radeon HD 5000-series goodies that make these new cards such an attractive prospect. Of course we're talking about Eyefinity triple-monitor gaming, bitstreaming high-def Blu-ray audio over HDMI, and DirectX 11 compatibility. On top of that the reference card is notably quieter than the 4890, although there is a compromise that has been made in the large size of the card.

Check out our full review here.

Radeon HD 5830: Bridging The Gap

"What?" you're asking. "There's another Radeon HD 5000-series card hiding under ATI's cap? That's right. In fact, there's more than one: the Radeon HD 5830 we're testing here today and a Radeon HD 5870 Eyefinity⁶ Edition board that'll surface in March. But we'll be looking at the six-display-capable model later. Today, with the GF100-based cards from Nvidia still missing in action, it's all about the Radeon HD 5830, purported to hit a price and performance point between the Radeon HD 5770 and Radeon HD 5850.

At this point, we should probably just refer to February 2010 as "Radeon 5000-Series Launch Month," because this is the third card to emerge in the last 28 days. It's also the most exciting of the trio. Nevertheless, if you missed the Radeon HD 5450 and Radeon HD 5570 stories, those are more entry-level discrete boards. This launch is aimed toward the folks with a little more money to spend on graphics, but not the $300+ it'd take to procure the company's higher-end offerings.

You see, before today, there was a big hole between the $160 Radeon HD 5770 and the $310 Radeon HD 5850. That's a gaping $150 price gap that fails to capitalize on the facts the ATI's previous generation fell off after the Radeon HD 4890 and Nvidia is only able to compete sub-$200 with GeForce GTX 260 or above $350 with the 285. The GeForce GTX 275 is essentially missing in action.

In its prime, the Radeon HD 4890 stood in for roughly $200, but good luck finding find one for sale anymore. The Radeon HD 4890 stock seems to have dried up over the last month.

Apparently, nature isn't the only thing that abhors a vacuum. AMD isn't too happy about it either, so it gave us the Radeon HD 5830 to fill the void. It emerges with a $239 launch price; promising, but certainly not set in stone, as the Radeon HD 5850 and 5870 showed us. Since the new 5830 is based on the high-end Cypress GPU that powers both of its higher-end predecessors, we have ambitious hopes for this card. In 1670, John Ray said, "Hell is paved with good intentions." So, let's make sure that this seemingly perfectly-placed product delivers the bang for your buck we'd expect.

The Radeon HD 5830 Architecture

There are a lot of graphics cards from AMD in and surrounding the new Radeon HD 5830's price point, so let's have a look at the relevant players:

There's so much information to process here. At first glance, the new Radeon HD 5830 seems to completely outclass the Radeon HD 5770 with far higher memory bandwidth and almost a third more shader cores. Indeed, the new card looks comparable to the powerful Radeon HD 5850. Yes, there's a 22% reduction in shader cores, but there's also a 10% increase in GPU core clock speed to help offset the deficit.

Things look exceptionally positive for the new Radeon HD 5830 until you notice its Achilles heel: a lack of ROPs compared to the Radeon HD 5850. And we thought we were so very close to having Radeon HD 5850-class performance at $240! Of course, this is precisely the reason AMD cut the number of render back-ends in half compared to the Radeon HD 5870 and 5850. To put things in perspective, the Radeon HD 5830 might be able to execute shader operations almost as fast as the 5850, but its ability to render pixels to the screen has been crippled to the same specification as the Radeon HD 5770. We can expect the new Radeon HD 5830 to perform between the Radeons HD 5770 and 5850, but probably closer to the 5770 or Radeon HD 4890.

Speaking of the 4890, the Radeon HD 5830 looks to outclass it on all fronts, except the ROP count. Remember though, the Radeon HD 4000-series have been repeatedly proven to demonstrate a sizable shader-versus-shader performance advantage over the Radeon HD 5000-series when it comes to raw game performance. As a result, we'd expect the new card to demonstrate performance comparable to the Radeon HD 4890.

A Familiar Graphics Processor

The Radeon HD 5830 does not sport a new GPU, it simply employs the Radeon HD 5870 "Cypress" processor, crippled to the new specification and dubbed Cypress LE. In fact, AMD's power usage specifications suggest that the Radeon HD 5830 will be based on the high-end Radeon HD 5870 card instead of the smaller 5850. And our reference card was built on a full-sized Radeon HD 5870 PCB, though we'll go into that in more depth on the next page. For now, here is a Radeon HD 5870 block diagram that shows how much of the GPU has been disabled.

We've been through the Radeon HD 5000-series architecture in detail in our Radeon HD 5870 launch article so I won't rehash the minutia. We will look at the differences in the new Radeon HD 5830 though: as you can see, the new Radeon HD 5830 has 14 SIMD engines, compared to the 5870's 20, but only half of the render back-ends. We can definitely see where the performance bottleneck will lie. Each of the 14 SIMD engines hosts four texture units and 16 stream processors, and each stream processor with five ALUs (AMD calls these stream cores). As a result, this GPU boasts 1,120 stream cores and 56 texture units.

Note that the GPU retains its four 64-bit memory controllers, even though the number of render back-ends has been halved from four to two. Each render back-end contains eight color ROP units, resulting in a total of 16 ROPs and a 256-bit memory interface.

As with the rest of the Radeon HD 5000 family, AMD dedicated a lot of its transistor budget to features like DirectX 11 and Eyefinity. Of course, all of these features carry over to the Radeon HD 5830. With the aforementioned launches this month, we've covered them ad infinitum, so to read more about these features please check the other Radeon HD 5000-series stories linked to at the end of this piece.

Radeon HD 5830: The Reference Card

At first glance, the reference Radeon HD 5830 appears to be a Radeon HD 5870. In fact, our test sample is physically identical to a Radeon HD 5870 reference card.

Note how the reference board dwarfs the Radeon HD 5850. I'm sure you might have heard this before, but size isn't everything. The specifications suggest that the Radeon HD 5830 will use more power than the 5850, despite its lower gaming performance. At over 11 inches long, the sheer size of this card might be a deterrent for many buyers. Even the already-large Radeon HD 4890 is an inch smaller.

The good news is that AMD included pictures of some of the planned boards from its third-party partners, and none of them appear as though the employ reference cooling. We aren't really able to make a judgment based on the stock art, but four of the cards seem to employ the longer PCB, while the XFX design is significantly shorter, potentially enabling a more compact version of the card.

As with the the Radeon HD 5870 and 5850, the Radeon HD 5830 requires two auxiliary six-pin PCIe power cables. This is a bit of a pity, since it would have been nice if only a single cable was required. But it's understandable given the full-sized Cypress GPU.

Our reference Radeon HD 5830 includes CrossFire-ready connectors. But we only had a single card to test for the launch, precluding dual-GPU benchmark results.

With two DVI outputs (in addition to DisplayPort and HDMI), the Radeon HD 5830 reference card gives us more connectivity than we could use in a typical environment. The need to acquire a DisplayPort-capable display for Eyefinity support might be an inconvenience for many enthusiasts, but it certainly beats having to sling two cards together for as many outputs. This will be one of the requirements imposed by Nvidia's Surround capability.

Yes, this GPU was created to be a fully-functioning Radeon HD 5870, but was subsequently crippled for its role as a Radeon HD 5830.

Test Setup And Benchmarks

When choosing cards to test against the new Radeon HD 5830, we looked to options in the same price league. The closest cards would be the Radeon HD 4890 and GeForce GTX 275, and both of these are just ridiculously unavailable right now. We had a Radeon HD 4890 sample on-hand, but were unable to find a GeForce GTX 275 at retail around our Canadian lab, underscoring the scarcity of these boards. We did manage to include a GeForce GTX 260, which might even prove a more relevant comparison, since this card can still be found from $200 to $220 online.

In this case, none of our test units are factory overclocked models. We left the cards clocked at their reference clock rates.

Benchmark Results: 3DMark Vantage And Crysis

3DMark sets us up for disappointment. You see, 3DMark is a shader-heavy benchmark, so the ROP-based bottleneck we were expecting plays a greater role in our real-world tests, performance might wind up falling close to the level of Radeon HD 4890.

Our first gaming benchmark demonstrates a literal tie between the Radeon HD 5830 and the increasingly-rare Radeon HD 4890. The only surprise here is that the 4890 wins by a slight margin, insignificant as it is. See how the Radeon HD 5850's increased ROP capacity launches it ahead of the pack?

At least the Radeon HD 5830 is showing an advantage over the 5770, but when you consider the $80 price difference between these cards, we aren't left all that impressed.

Benchmark Results: Call Of Duty: Modern Warfare 2

Call of Duty: Modern Warfare 2 delivers the same general scenario that Crysis did, with the Radeon HD 5830 achieving a virtual tie with the Radeon HD 4890. Even with 4xAA enabled in this title, the Radeon HD 5770 holds up well. After all, this is a very forgiving game engine, despite the great visuals.

Benchmark Results: Resident Evil 5

Resident Evil 5 continues the status quo, showing us that the new Radeon HD 5830 can be mistaken for a Radeon HD 4890 if all you're concerned about is raw gaming performance. The Radeon HD 5850 certainly looks worth the extra money, though, and the 5770 looks like an increasingly-attractive value.

Benchmark Results: World In Conflict: Soviet Assault

The World in Conflict results don't show us anything new. It's surprising how well the Radeon HD 5770 can keep up to the more powerful cards when you consider its 128-bit memory interface, even with anti-aliasing applied.

Benchmark Results: Fallout 3

Fallout 3 demonstrates that, sometimes, the Radeon HD 4890 will show more than a couple FPS lead over the 5830. Other than that, there's nothing new to learn here.

Benchmark Results: Tom Clancy's H.A.W.X.

The only information of note here is that the GeForce GTX 260 seems to be limited by its 896MB of video RAM compared to its competitors and their full 1GB frame buffers. We're likely seeing this limitation due to the 8x anti-aliasing mode we've used to benchmark this title. We chose 8x AA because 4xAA doesn't stress the graphics cards as much.

Benchmark Results: DiRT 2 And DirectX 11

DiRT2 lets us explore DirectX 11 performance, and it looks like the Radeon HD 5830 performs much closer to the Radeon HD 5770 than it does the Radeon HD 5850 in this particular title.

Of course, it will be interesting to see how upcoming DirectX 11-class games change the story. Aliens Vs. Predator, for instance, is a title we'll be reviewing in the near future. This one one promises to showcase tessellation, and we're anxious to see what that does to performance.

Benchmark Results: Left 4 Dead And AA/AF

Left 4 Dead isn't particularly hard on the graphics card, so it gives us an opportunity to isolate the anti-aliasing and anisotropic filtering ability of these products. Thanks to the GDDR5 memory and full 256-bit memory interface, here is a benchmark where we can see the new 5830 stretch its legs a little, although the Radeon HD 5850 continues to hold a strong lead.

Power, Temperature, And Noise Benchmarks

There are a few surprises in the power usage tests. The new Radeon HD 5830 does use a few more watts than the Radeon HD 5850 at idle, likely due to the Radeon HD 5870 PCB on which our reference card is built. But, under load, the new card doesn't seem to be as stressed.

This is the opposite result that we expected after looking at AMD's specifications, which suggest the 5830 uses less power at idle and more under load. We're also surprised to see the Radeon HD 4890 drawing less power than the GeForce GTX 260 sample we have on-hand. The Asus ENGTX260 Matrix is a non-reference card, so we're not sure if optimization on Asus' end has anything to do with the result. However, we expected the Radeon HD 4890 to pull more watts than its competition.

Notice how little power the Radeon HD 5770 uses in comparison, an impressive result for a card that performs relatively closely to the new Radeon HD 5830.

All of these temperatures are well within reasonable limits. It is notable that the new Radeon HD 5830 produces much less heat than the other cards, despite the fact that it is saddled with a reference cooler. Then again, it makes sense when you consider that this reference cooler was designed to keep the powerful Radeon HD 5870 cool enough under load. Also note that the Asus GeForce GTX 260 is brandishing a powerful non-reference cooler and is achieving very low temperatures.

The Radeon HD 5830 generates surprisingly little noise, probably because the Radeon HD 5870 reference cooler doesn't need to spin very fast in order to keep things cool. Here, the Radeon HD 5830 offers a significant benefit compared to the notably-noisier Radeon HD 4890.

Conclusion

With the introduction of its Radeon HD 5830, AMD has now filled the most obvious price gap in its DirectX 11-capable lineup. The company has also given us another powerful card that takes the torch from the Radeon HD 4890, as it is sent out to pasture.

Incredibly similar gaming performance makes comparisons between the Radeon HD 5830 and Radeon HD 4890 easy. The similarities don't end there, as both cards were launched at similar price points. Almost certainly, given enough time and (hopefully) some competition, the Radeon HD 5830 will drop in price and deliver the same price/performance ratio that the Radeon HD 4890 has spoiled us with. And let us not forget that the 5830 delivers more than game performance alone. It uses half the idle power of the Radeon HD 4890, notably less power under load, and supplies us with all of the Radeon HD 5000-series goodies that make these new cards such an attractive prospect. Of course we're talking about Eyefinity triple-monitor gaming, bitstreaming high-def Blu-ray audio over HDMI, and DirectX 11 compatibility. On top of that the reference card is notably quieter than the 4890, although there is a compromise that has been made in the large size of the card.

What's the worst thing we can say about the new Radeon HD 5830? Well, its launch price is $80 higher than most of the Radeon HD 5770 cards in the wild, and only $60 less than most of the Radeon HD 5850s available. Based on relative performance, this relationship should probably be reversed, as the new card performs much closer to the 5770 than the 5850. On top of that, we've been spoiled by the Radeon HD 4890 delivering a solid level of performance at $200 for months now, so spending an extra $40 for performance at the same level doesn't inspire us. The market dictates what it will bear, but with Radeon HD 4890s disappearing, it won't be possible to make comparisons for long.

Pricing concerns aside, the Radeon HD 5830 is a strong performer armed with the universal feature set we've seen from all Radeon HD 5000-series cards thus far, specializing in quiet operation and moderate power usage. It ups the ante compared to the Radeon HD 5770 with almost twice the memory bandwidth; this can be a solid boon in some scenarios.

AMD is undoubtedly making a killing with its monopoly on DirectX 11-class cards, but Nvidia hasn't thrown in the towel yet. We can only hope that its upcoming answer to the Radeon 5000-series (expected to begin rolling out at the end of March) will add some flavor and choice to the market. Until then, AMD has an unbeatable selection of cards in its hand. You might even call it a royal flush, and if the dual-GPU flagship Radeon HD 5970 is a performance ace, this Radeon HD 5830 is its jack.

If you want to know more about the features defining the Radeon HD 5000-series cards, please check our other reviews below (particularly the Radeon HD 5870 article, which was the first in the series).

Radeon HD 5970
Radeon HD 5870
Radeon HD 5850
Radeon HD 5770
Radeon HD 5750
Radeon HD 5670
Radeon HD 5570
Radeon HD 5450

Monday MSI revealed its new 16-inch gaming notebook, the GE600, which uses Intel's HM55 chipset and 32nm Core i5 processor, and AMD's ATI Radeon HD5730 discrete graphics (with 1 GB DDR3 VRAM).

The portable rig sports a 720p HD webcam, a built-in sub-woofer (really?), and SRS Premium Sound. The GE600 is also decked out with a shiny coat of black, highlighted by a gold palm rest.

As for other specs, the gaming notebook offers a Windows 7 Home Premium, two DDR3 RAM SO-DIMM slots, a 16-inch HD TFT LCD screen with 1920 x 1080 resolution), and SATA HDD space ranging from 250, 320, to 500 GB. There's also a 4-in-1 card reader (XD/SD/MMC/Memory Stick), Bluetooth, Intel's 802.11 a/b/g/draft n, HDMI out, and more.

MSI said that it is also throwing in two "theater-class" speakers and its exclusive ECO Engine Power Saving Technology. The laptop will come with a six-cell battery, however gamers can update to the nine-cell version.

"To satisfy customer needs, MSI 16" entertainment model - GE600 - was developed to introduce the most advanced software/hardware devices characterized by all-new exterior design, HD webcam, sub-woofer and leading SRS Premium Sound audio, all of which enables customers to enter a new era of mobile video/audio enjoyment experiences," said Sam Chern, MSI Notebook Marketing Director.

Outside the specs and all the promotion, MSI didn't offer any pricing or an exact release date on this laptop.

VIA Technologies said that it's sporting a few products at the Digital Signage Expo 2010 trade show held next week at the Las Vegas Convention Center. One of the products mentioned is the company's upcoming S3 Graphics Chrome 5400E x2, a dual-GPU graphics card aimed to take on AMD's ATI Eyefinity technology.

"The S3 Graphics 5400E x2 is a highly integrated and flexible video card that features dual graphics and multimedia processors," the company said. "Boasting extreme hardware acceleration of the latest HD video codecs, the S3 Graphics 5400E x2 uses S3 Graphics' PanoChrome technology to provide up to four independent HD video streams at resolutions of 1080p on up to eight displays, employing a variety of display modes and configurations."

The company's press release provided a few more details, indicating that the card provides built-in Genlock support for synchronized source timing, and support for Span, Extended, and Clone view configurations. The hardware-accelerated HD video codecs include H.264, VC-1 and WMV-HD. Unfortunately, there's no real hardware-based info such as its core clock speed and on-board memory amount.

VIA said that it will offer a taste of the dual-GPU graphics card during the show, and will eventually provide samples to ODM customers in the USA sometime Q2 2010 or later. More information about what VIA is bringing to the trade show next week can be seen here.

Looking for the best bang-for-buck video card to fuel your budget sensitive PC gaming habit? Then AMD may have the answer for you. AMD today introduced the ATI Radeon HD 5570 graphics card, packing full support for DirectX 11, as well as ATI Eyefinity multi-display technology and ATI Stream capabilities.

"AMD recognizes that small form factor PCs are becoming more popular and low profile graphics upgrade options have been limited to date," said Matt Skynner, vice president and general manager, AMD Graphics Division. "Customers purchasing small form factor PCs are looking for improved performance while gaming, watching HD video or working with the latest productivity applications.  The ATI Radeon HD 5570 graphics card delivers all of this at a price that won't break the bank."

This latest offering completes a full DirectX 11 refresh of AMD discrete graphics products – a milestone the graphics division has to be pleased about.

From our own hands-on with the card, we found it to be an excellent value at just $80.

Consider that this is a half-height reference card, able to transform even the smallest systems into viable gaming machines. Power usage is extremely low for the performance offered, and no auxiliary power connector is needed. And yet, the Radeon HD 5570 manages playable frame rates in every one of our game tests at 1680x1050 (and sometimes 1920x1200). Triple-monitor Eyefinity gaming could be viable in less-demanding titles, such as World of Warcraft, at a price substantially lower than the Radeon HD 5670. Just make sure the board you buy has a DisplayPort output first.

All of these features make it easier to recommend the Radeon HD 5570 over the slightly less expensive (but notably slower) Radeon HD 5450.

Check out our full review here.