The first half of this year felt like Nvidia was taking a time-out or a nap, leaving both its third-party board vendors and customers impatient (BFG, of course, paid the ultimate price). It seemed the company was content to keep creatively renaming its desktop graphics chips, and thus sticking with its then-current portfolio.
Meanwhile, AMD launched its entire line of DirectX 11-compatible Radeon HD 5000-series cards, literally from top to bottom.
Finally, after several delays, Nvidia brought its own DirectX 11 GPU to market, officially called GF100 and based on the Fermi architecture.
The first cards to employ the new design were called the GeForce GTX 480, GeForce GTX 470, and GeForce GTX 465, and all of them were aimed at the gaming enthusiast. Now, Nvidia is introducing a workstation-class card belonging to the Quadro line. Again, AMD enjoys a bit of a lead here as well, having already updated its FirePro line of professional graphics cards in late spring.
That sets the stage for today’s comparison in which we will pit Nvidia’s new Quadro 5000 against AMD's flagship ATI FirePro V8800, comparing both cards to previous-generation parts as well. Since some of our readers may not be familiar with workstation graphics, allow us to provide a frame of reference by comparing them to their mainstream gaming brethren.
The Quadro 5000 is based on Nvidias GF100 GPU and contains 352 shader cores, making it more or less a twin of the GeForce GTX 465. The main differences between the two relate to the memory subsystem and the clock frequencies. AMD's FirePro V8800 is built around the company's Cypress XT (RV870), the same GPU powering its Radeon HD 5870.
So, going into this comparison, what should our expectations be? Comparisons between the gaming versions of these cards show them going head-to-head in many cases, with AMD taking the lead in some cases and Nvidia in others, depending on the game, benchmark, and settings. In the past, this sort of pattern has translated quite well into the workstation world. Will this hold true here as well, or will we see a clear winner emerge?
When it unveiled its Fermi architecture, Nvidia repeatedly stressed that this design was not developed expressly for the gaming market. Instead, the company strove to create a design that could also target the workstation segment, as well as HPC markets. Today‘s test will also show whether Nvidia achieves this goal.
Before we dive into the benchmarks, we’ll take a closer look at the contenders, sizing up Nvidia’s Quadro 5000 and the ATI FirePro V8800. We’ll also consider their respective fields.
| Model | GPU | Fab | Mainstream Equivalent | Memory | DVI | DisplayPort | 3-Pin Stereo | Max Power |
|---|---|---|---|---|---|---|---|---|
| ATi FirePro V8800 | RV870 Cypress | 40 nm | Radeon HD 5870 | 2048 MB GDDR5 (256-bit) | 2 (Adapter) | 4 | Yes | 208 W |
| ATi FirePro V8700 | RV770 | 55 nm | Radeon HD 4870 | 1024 MB GDDR5 (256-bit) | 1 | 2 | Yes | 151 W |
| Nvidia Quadro 5000 | GF100 Fermi | 40 nm | GeForce GTX 465 | 2560 MB GDDR5 (320-bit) | 1 | 2 | Yes | 152 W |
| Nvidia Quadro FX 4800 | GT200 | 65 nm | GeForce GTX 260 | 1536 MB GDDR3 (384-bit) | 1 | 2 | Yes | 150 W |
| Model | Memory Bandwidth | DirectX | OpenGL | Shader Model | Core Clock | Memory Clock | Pixel & Vertex Processing |
|---|---|---|---|---|---|---|---|
| ATi FirePro V8800 | 147.2 GB/s | 11.0 | 4.1 | 5.0 | 825 MHz | 1150 MHz | 1600 SPUs |
| ATi FirePro V8700 | 115.2 GB/s | 10.1 | 4.1 | 4.1 | 750 MHz | 900 MHz | 800 SPUs |
| Nvidia Quadro 5000 | 120 GB/s | 11.0 | 4.1 | 5.0 | 513 MHz | 1500 MHz | 352 SPUs |
| Nvidia Quadro FX 4800 | 76.8 GB/s | 10.0 | 3.1 | 4.0 | 600 MHz | 800 MHz | 192 SPUs |
New Applications
Workstation graphics cards of this caliber are usually employed by engineers. Companies like Autodesk, Dassault and Mental Images offer specialized software solutions. Less-powerful models are a good fit for less-demanding applications in fields like digital content creation and desktop publishing.
As a result of the increasing popularity of 3D movies, demand for workstation cards has increased dramatically in Hollywood and its production studios. In the case of films aimed at a theatrical release (as opposed to straight-to-DVD production) the producers often use a resolution that is four times that of full HD. These companies need a quick way to check how their 3D models will appear to the audience. Since time is money, they obviously want to see the results as quickly as possible, and can’t afford long rendering times every time one of the parameters of a model is altered. One such production company is The Foundry, which also produced several scenes for last year’s blockbuster Avatar.
Workstation cards are also very beneficial for video enhancement. For example, MotionDSP offers a software solution that acts as a retroactive image stabilizer for shaky home videos. One solution used by the police and the military is called Ikena. There is even a freeware variant available to home users called vReveal that also offers image stabilization, but has some other tricks up its sleeve as well.
Current developments in the medical sector are interesting as well. Here, powerful workstation solutions like Nvidia's Quadro are used in real-time 4D imaging. Until recently, imaging systems, such as ultrasound and CAT scans, only provided moving images in 2D. 3D images, on the other hand, were always snapshots. That is different today. Medical equipment able to show the beating heart and the flow of blood through its chambers usually rely on workstation-class graphics cards like the ones we are looking at in this review.
Nvidia is shortening the branding of its workstation cards. While the previous generations were dubbed Quadro FX, the new models will simply be called Quadro, followed by their model number. The FX suffix that many buyers associate with “special effects” has been dropped completely. AMD made a similar move last year, renaming its FireGL line to FirePro. The “GL,” which referenced the OpenGL graphics language, has been switched out for Pro, connoting “professional” instead.
The Quadro 5000, built by Nvidia and distributed by PNY, is considered the secret flagship of the Quadro family. Sure, there’s also the Quadro 6000, which is equivalent to the desktop GeForce GTX 480, just with an overabundance of memory. Of course, this model will surely find buyers as well, but the target audience is much smaller. That’s partly due to a steep price, but also because of there is only a limited number of applications out there that would benefit from such a card. We saw the same thing with the previous-generation Quadro FX 4800 and FX 5800.
So what kind of graphics card is the Quadro 5000, really? Well, under the hood, we find a GF100 GPU with 352 shader cores enabled. That makes it equivalent to the GeForce GTX 465 found in the desktop space, albeit a slightly chubbier and slower one. After all the, Quadro 5000 comes with more than twice as much memory as the GeForce GTX 465, but runs at lower clock speeds. The following table gives you a side-by-side comparison of the two cards.
| Specifications | Quadro 5000 | GeForce GTX 465 |
|---|---|---|
| Chip | GF100 / Fermi | GF100 / Fermi |
| Memory | 2560 MB | 1024 MB |
| Core Clock Rate | 513 MHz | 607 MHz |
| Shader Clock Rate | 1026 MHz | 1215 MHz |
| Memory Clock Rate | 1500 MHz | 1600 MHz |
In all, Nvidia launched a total of five products for the workstation and server markets.
Scalable visualization system:
- Quadro Plex 7000, 12 GB memory, 896 CUDA cores
Desktop workstation:
- Quadro 6000, 6 GB GDDR5 memory, 448 CUDA cores
- Quadro 5000, 2.5 GB GDDR5 memory, 352 CUDA cores
- Quadro 4000, 2 GB GDDR5 memory, 256 CUDA cores
Mobile workstations:
- Quadro 5000M, 2 GB GDDR5 memory, 320 CUDA cores
DisplayPort is the graphics card industry’s new favorite connector, since it guarantees high scalability for upcoming display solutions. There’s just one catch. How do you establish a new connector if the monitor makers aren’t willing to play along, and the majority of users have only just made the switch from VGA to DVI? Many folks may not even quite know what to make of HDMI yet, much less DisplayPort.
Nvidia’s approach is a cautious one, and while the company equips the Quadro 5000 with two DisplayPort connectors, it also provides a single dual-link DVI output. However, unless your monitor is already compatible with DisplayPort, you’ll still need to buy additional adapters if you’re planning to use a multi-monitor setup.
The memory system has also undergone an evolutionary change in that it now supports ECC (error correction code), making the Quadro 5000 the first card with this capability. The technique is not necessarily all that relevant to image processing. However, it is of great importance in medical analysis, financial computation, and cluster-based configurations. Even small single-bit errors can have a tremendous impact on the final result. ECC allows the graphics card to detect and correct this type of error, just like server and workstation motherboards can with system memory. The downside is that it results in a performance penalty. By default, Nvidia deactivates this feature in its drivers.
Nvidia also provides a 3-pin DIN port for use with 3D shutter glasses on the card’s backplate. The company already has compatible wireless solutions in its product portfolio as well.
Feature-wise, Nvidia is competitive with AMD once again. Shader Model 5, DirectX 11, OpenGL 4.1, and OpenCL 1.0 are all finally supported in this generation of GPUs after several delays. Special solutions like Framelock, Genlock, and Serial Digital Interface required by the broadcast industry are also provided by this card.
AMD's ATI FirePro V8800 succeeds the FirePro V8700. Going only by the number of shader cores the V8800 is able to muster, you’d think there should be no competition. At first glance, the GPU’s 1600 shaders would indicate impressive performance potential. By comparison, Nvidia's 352 CUDA cores sound downright paltry. However, as our benchmarks will show, it’s not just about how many shaders you’ve got, it’s how you use them. At the end of the day, whether the cores are called CUDA or Stream says nothing about their capabilities.
As mentioned, the FirePro V8800 is closely related to the Radeon HD 5870. Both are built around the Cypress XT GPU, also known as RV870. Like Nvidia, AMD artificially hobbles its Radeon-class desktop cards through the driver, ensuring that only FirePro models will offer top performance in workstation scenarios. While a Radeon may seem like the less expensive choice, the upshot is that it’s a less-than-ideal card for a workstation.
How closely are these two cards related, then? Superficially, the Radeon HD 5870 and the FirePro V8800 are easy to tell apart simply by the product names printed on their shrouds and the model numbers on the PCB. Also, AMD has opted to rely completely on DisplayPort connections for its professional model, while the Radeon HD 5870 still comes with two DVI outputs and an HDMI connector as well. However, even a practiced eye would have trouble telling the two cards apart based only on their board layouts.
As expected, the ATI Fire Pro V8800 uses lower clock speeds than its desktop sibling. Again, the following table provides an easy comparison.
| Specifications | FirePro V8800 | Radeon HD 5870 |
|---|---|---|
| Chip | Cypress XT (RV870) | Cypress XT (RV870) |
| Memory | 2048 MB | 1024 MB |
| Core Clock Rate | 825 MHz | 850 MHz |
| Memory Clock Rate | 1150 MHz | 1200 MHz |
ATI has launched several workstation cards this year. These include the following models.
- ATI FirePro V8800 (2 GB GDDR5, Cypress XT / RV870)
- ATI FirePro V7800 (2 GB GDDR5, Cypress Pro / RV870)
- ATI FirePro V5800 (1 GB GDDR5, Juniper XT / RV840)
- ATI FirePro V4800 (1 GB GDDR5, Redwood XT / RV830)
- ATI FirePro V3800 (1 GB GDDR3, Redwood XT / RV830)
Our FirePro V8800 sample completely eschews DVI. Instead, the card sports four DisplayPort connectors, allowing four displays to be used in parallel. As on the desktop, this feature is referred to as EyeFinity.
Considering the small installed user base, a DisplayPort-only design is a rather bold move. After all, most folks still have a monitor with a DVI cable sitting on their desk. There‘s no need to worry, though, as ATI includes two active single-link adapters with the card that can translate the DisplayPort output to a DVI signal. Since DisplayPort and DVI are not compatible on the physical protocol level, these need to be active adapters, with a circuit translating the signal from one protocol to the other.
As a side note, DVI-to-HDMI adapters (or vice versa) don’t need an active circuit, since in both cases, the video signal is based on the PanelLink protocol. Instead, you just need an adapter that contains connections linking the various pins for each plug correctly.
The FirePro V8800 also features a 3-pin connector for stereoscopic 3D glasses. Again, that doesn’t necessarily mean the glasses have to be connected by wire. Instead, a little transmitter that synchronizes the stereoscopic images will do just as well.
AMD can put checkmarks in all the important feature boxes, including Shader Model 5.0, DirectX 11 and OpenGL 4.1. The company knows just as well as Nvidia that modern graphics processors are capable of much more than just rendering pretty pictures. The high transistor count and the many shader cores make current GPUs an attractive choice for scientific calculations. While Nvidia is actively promoting its proprietary CUDA API for GPGPU work (general purpose graphics processing), AMD has all but dropped its Stream API, choosing instead to back the OpenCL standard. While that move is laudable on paper, it’s not like developers are falling over themselves to create software that makes use of OpenCL (yet).
Nvidia, for its part, regularly promotes CUDA at universities, scientific institutions, and to its partners in various industrial sectors in order to inspire developers to use its programming language. And yes, that includes financial support as well.
AMD offers an add-in card for specific usage scenarios called the FirePro S400. Production companies working in the broadcasting sector need Framelock and Genlock when working across several monitors in parallel. The S400 offers this type of synchronization.
The FirePro V8800 is rated with a TDP of 208 W and satisfies its power requirements through two 6-pin auxiliary power connectors. The card also comes with a CrossFire connector that allows two V8800s to be used in parallel.
When we reviewed the previous generation of workstation cards, we were faced with an unsatisfying situation. Although Widows Vista had been on the market for quite a while at that point, we were forced to fall back on Windows XP for testing. After consulting with several contacts in the CAD sector, the reason became clear, though. Due to instability and security concerns, only very few professionals were willing to make the switch to Windows Vista. Over the past year, this situation has changed.
Windows 7 has received a much warmer welcome than Vista. On top of that, most professional users are making the big switch, upgrading not only to Microsoft’s newest OS but choosing its 64-bit version at the same time. This promises better performance, as well as improved memory management.
At any rate, we‘re quite happy to move our OpenGL test platform to the new OS so that we’ll be using the 64-bit version of Windows 7 Enterprise in this review. There are already some scripts for 64-bit apps available for download at www.spec.org, although they seem to be appearing only very slowly.
| System Hardware | |
|---|---|
| CPU | Intel Core i7-965 Extreme |
| Motherboard | Asus P6T (rev 1.01G) with Intel X58 Chipset |
| System Memory (RAM) | 6 GB (3 x 2 GB) Corsair DDR3 1600 MHz @ 8-8-8-24 |
| Optical Drive | Samsung SH-D163A , SATA 1.5 Gb/s |
| Power Supply | Zalman, ATX 2.01, 510 W |
| Hard Drive | Western Digital WD1500ADFD 150 GB |
| System Software And Tests | |
| Operating System | Windows 7 Enterprise 64-bit |
| DirectX | DirectX 11.0 |
| Intel Chipset Software | Version 9.1.2.1007 |
| ATI FirePro Driver | Catalyst 8.723 |
| Nvidia Quadro Driver | Quadro 258.98 |
| SPEC Benchmark Settings | Application settings according to SPEC Project Group Rules, driver using application optimizations if available |
| SPEC Benchmarks Being Used | SPECViewperf 11 SPECapc Autodesk Maya 2009 v1.0 SPECapc Newtek LightWave 9.6 SPECapc Autodesk 3D Studio Max 9 1.2 |
Autodesk 3D Studio Max is widely used among professionals. Nvidia offers a special performance driver for this application that needs to be installed in addition to the normal one. The extra step is worth it, though. Without it, the Quadro’s lead would not have been as pronounced. AMD also used to offer a tuned driver for 3DS Max. However, its functionality has been folded into the main release.
Unfortunately, SPEC doesn’t offer a script for the newest version of 3D Studio Max, which is why we are using the older one. The results show Nvidia clearly dominating this graphically-intensive benchmark. In this case, even the older Quadro FX 4800 is able to outpace AMD’s current flagship, the V8800.


In the 2009 version of the SPECapc Maya benchmark, this year’s cards are faster than last year’s models. Interestingly, while the Quadro 5000 beats the FirePro V8800, the situation is reversed in the previous generation, with AMD pushing ahead of the Nvidia card.


The SPECapc benchmark for Newtek’s Lightwave 9.6 produces mixed results. In multitasked workloads, Nvidia can eke out a slim lead, while AMD carries the interactive category.


Recently, SPEC released version 11 of its OpenGL benchmark. Thanks to a greatly simplified user interface, even novice users shouldn’t have any trouble using the test suite. Due to its synthetic nature, Viewperf is a perfect target (unfortunately) for driver optimizations. A well-rounded review should always take these results in the greater context of real-world applications
At any rate, the Quadro 5000 wins every single discipline that the test suite hast to offer. In light of these results, we think it’s safe to assume that AMD is currently working on an updated version of its driver that could minimize or even close the performance gap between the new Quadro and its own FirePro V8800. At least that would seem like the logical conclusion after looking at the previous benchmark results we collected in Maya, 3ds Max, and Lightwave. Thus, it’s quite possible that we will be able to update these charts sometime soon.







Looking at our benchmark results, it’s clear which card comes out on top: Nvidia’s Quadro 5000 is superior to AMD’s FirePro V8800 in almost every benchmark, usually by a clear margin. Based on their mainstream equivalents, we wouldn’t have expected such a great performance disparity between these two graphics boards. After all, AMD's offering is built on an excellent modern design, though it seems to have trouble living up to its performance potential in a workstation environment.
Indeed, we don't think that it's the hardware to blame. After all, the Radeon HD 5870 is faster than the GeForce GTX 465. Rather, it's AMD’s driver at fault. While they are stable, they are apparently not very well optimized yet, holding the V8800 back. This isn’t just the case compared to Nvidia's current flagship. The V8800 should have been able to outpace its own predecessor by a much larger margin. AMD's best approach would be to commit more resources to its driver team to rectify the situation.
Nvidia’s newest software package grants the Quadro 5000 a clear lead in the synthetic benchmark SPECviewperf 11. While the Nvidia card also comes out ahead of the AMD board in real-world benchmarks, the margins are not as wide as the synthetic suite would imply. 3ds Max is the one exception, with the previous-gen Quadro FX 4800 beating not only its contemporary FirePro V8700, but the new V8800 as well. We’re happy to see that Nvidia has optimized its drivers for performance in several applications, which is something that AMD has been implementing in its own drivers for a while now. As a result, there is less need to manually adjust software-based settings when using an Nvidia card.
Looking at the two competitors from a "value for the money" perspective, Nvidia certainly appears to be the premium choice, both where performance and cost are concerned. Its faster card is naturally more expensive. Meanwhile, AMD prices its cards at a level that is in line with its capabilities. In other words, the AMD cards are less expensive and still offer respectable performance, but they’re not the fastest models out there.
Depending on the route you go, pricing may not be as much of an issue. While workstation graphics cards can be bought separately, the majority are sold as part of a preconfigured workstation computer from companies like Dell (Precision line) or Hewlett-Packard (z800), where the customer can choose the components. In the context of an entire workstation, the cost of a graphics card is only a small part of the final sum.














