Related articles

• "Nvidia's got to face reality: If they want to grow, they have to get into other markets," remarked Jim McGregor, editor of In-Stat's Microprocessor Report. "I doubt ATI is going to get out of offering the high-performance chipsets, so I wouldn't give the market to Nvidia just yet. But whether or not they will actually be able to garner additional revenue from Intel, I wouldn't necessarily give them that [deal] either." "Probably the knee-jerk reaction is...Nvidia should be rushing into Intel's arms. But I don't think that's quite the case yet." Jim McGregor, editor, In-Stat Microprocessor Report The way McGregor sees it, there may be nothing certain that emerges from this deal for Nvidia and Intel. Just because Intel doesn't have a presence in performance graphics today, he said, doesn't mean it won't have one tomorrow, and Nvidia doesn't necessarily have to play a role in that. Intel is as interested in building new platforms as AMD; but with one consumer platform already garnering success - Centrino - Intel has already proven it can meet the demands of its ODM customers without relying on anyone else's intellectual property. In other words, it can take the dance floor on its own and still command the floor. So Nvidia's next move with regard to Intel could depend on whether ATI appears to cede its share of the high-end market. And if ATI does that, McGregor believes, then it's cutting off its own lifeline. "It's a lot easier to dumb something down than it is to build it up," he told TG Daily. So do I think ATI's going to get out of the high-end graphics market? Absolutely not. If they do, they're worthless to AMD...just because that technology that they continue to develop is critical for them to remain competitive from a graphics standpoint. If they gave up that market, I'd say AMD wasted their money." If Intel continues to go it alone in the platform department, it might not leave Nvidia in the cold, believes IDC semiconductor analyst Shane Rau. In a sense, he told us, Nvidia already has a platform: SLI. It's a concept that requires two or four (maybe more) Nvidia GeForce GPUs, and an Nvidia nForce chipset. "What is the fate of Nvidia long-term?" posed Rau, beating us to the punch. "Initially, it doesn't look very good, because they're being cut out. Probably the knee-jerk reaction is, 'Oh, they've gotta find a partner, they've gotta get bought, too!' And Nvidia should be rushing into Intel's arms. But I don't think that's quite the case yet. I think there is an argument that these companies can maintain independence based on their ability to provide some independence and some third-party-based differentiation." Rau believes that a technology platform need not necessarily be comprised of all one brand of silicon, assembled in one prescribed manner. If done properly, a platform can spell out exactly what areas a company can innovate, and still innovate with the system: case in point, the whole PC expansion bus thing to begin with. "OEMs who make the systems want to be able to differentiate," he told us. "They don't want to have all the same silicon in their system. They want to be able to differentiate on something." "Alienware, Falcon Northwest...don't want to have the same processor, the same chipset, the same graphics, the same networking components as all the mainstream PC OEMs." Shane Rau, semiconductor analyst, IDC This could put Nvidia in a strangely advantageous position. Specifically, with AMD and ATI jointly going after ODM customers, responding to what they say they want - a standardized platform - Nvidia could easily go after the same customers and others, responding to something eise they say they want: choice. And for the high-end, performance, and enthusiasts' market segment, Nvidia could end up being the company that exemplifies choice. In so doing, SLI could very conceivably become the platform...of choice. "I think differentiation and segmentation kind of go hand in hand," remarked IDC's Rau. "I think segmentation is partially driven by the fact the market gets so big that not everybody can do everything well, so they pick something that they can do well and they do it so well that it becomes a niche, or eventually a segment unto itself." He pointed out specific examples: "Alienware, Falcon Northwest - small, but differentiated PC companies who decided to address high-end PC enthusiasts, and they want to continue to have that differentiation. They therefore don't want to have the same processor, the same chipset, the same graphics, the same networking components as all the mainstream PC OEMs. They want to mix and match; they want to have the best processor they can get - for their purposes, from an Intel or an AMD. Then they've got to find the best chipset they can get, and that may not be from Intel or AMD, it may be from Nvidia. So you can start to see an argument that Nvidia could play both sides of the fence by offering some ability to differentiate, to innovate, within an Intel platform or an AMD platform." The Nvidia platform today, said Rau, "doesn't have a central processor unit, but it does have GPU technology, of course, and now [Nvidia] has a chipset business. They have the SLI technology, which requires two or more Nvidia GPUs and an Nvidia chipset put together. This is a platform approach, and they've been able to do this successfully; they pressured ATI to come up with a similar solution [Crossfire]. So if you want to have this technology, you couldn't do it with Intel silicon. You couldn't do it with AMD silicon alone. You needed to go to Nvidia for this. So I think that is an example of how Nvidia can survive by innovating and providing differentiation for its OEM customers."

• Full Review NVIDIA's new GeForce256 'GPU'

  The first time we heard about NVIDIA's mysterious 'NV10'-project is now more than 1.5 years ago. Still NVIDIA took time until this 'super-chip' was eventually presented to the public. It's only a month ago that 'GeForce256' was finally unveiled to us, and NVIDIA didn't lose any time to make it available to everyone soon after the announcement, although the schedule got delayed by the Taiwanese earthquake. Creative Labs were even faster than NVIDIA, they released their first boards already last week in Asia and I doubt that it will take much longer until you will be able to buy the '3D Blaster Annihilator' in the US and Europe too. I only wish that the names of those 3D-products wouldn't always have something to do with mass-murder. Since 'Napalm' I'm only waiting for a 3D-product called Hiroshima, Agent Orange or Pearl Harbor. Anyway, to get back to the topic, what's the news about GeForce? Well, most of you will certainly have read several articles about its architecture elsewhere, and while we at Tom's Hardware don't claim to fame with regurgitating presentations or white papers, I will try and keep my summary of GeForce256 short. For those of you who require more information I deeply suggest visiting NVIDIA's website , where you can find all the white papers and presentation material you need to keep dreaming about the GeForce GPU all night long. Download image in PNG format This is the GeForce256 reference board with single data rate SDRAM, actually made by Creative Labs and thus identical to '3D Blaster Annihilator'. Download image in PNG format The GeForce256 reference board with DDR-SGRAM looks like this. The Features The most important thing about GeForce is that NVIDIA doesn't want it to be called '3D-chip' anymore. GeForce is called a 'GPU' for 'Graphics Processing Unit', and this name was chosen to give it the right honors in comparison with the all-important 'CPU'. Most of us don't realize that the complexity of today's graphics chips is at least on par with any high-end PC-processor, while the price point of 3D-chips is rather mediocre compared to a CPU. GeForce ensembles no less than 23 million transistors, which is in the same range as AMD's Athlon or Intel's upcoming Coppermine processor. Intel or AMD take more than 10 times more money for one, which makes you wonder why the 3D-chip industry is still alive. The 23 million transistors in GeForce are well spent, since it's the first 3D-chip that includes a transform and lighting ('TNL') engine. This engine is the real reason for the name 'GPU', because it adds a huge amount of computing power to GeForce, which makes it a lot more than just a chip that displays 3D-graphics. Transform & Lighting Engine 'Transforming' is the very FP-calculation intensive task to 'transform' the 3D-scene with all its objects, called 'world-space' into the 'screen-space' that we are looking at. 'Lighting' is pretty self-explanatory, it represents on optional stage in the 3D-pipeline that calculates the lighting of objects in relation to one or several light sources. Lighting is just as transforming a pretty FP-calculation intensive task. Both tasks used to be executed by the CPU, putting rather heavy strain on it. The effect was that the 3D-chip was often in the situation that it had to wait for the CPU to deliver data (e.g. CPU-limited 3D-benchmarks) and that game developers had to restrict themselves to less detailed 3D-worlds, because heavy polygon usage would have stalled the CPU. GeForce is supposed to put an end to this dilemma, it can take the strain off the CPU, can keep the 3D-pipeline from stalling and allows game developers to use much more polygons, which automatically results in greatly increased detail. At the same time the CPU can dedicate itself to a more complex game AI or more realistic game physics. NVIDIA claims that GeForce can transform, clip and light 15-25 million triangles/s, and somewhere else I read it's at least 10 million triangles/s. High Fill Rate Well, even GeForce is not free from having to obey the classic demands in 3D-gaming, and one of those is asking for as much fill rate as possible. NVIDIA once claimed 500 Mpixels/s, now it's obviously shrunk to 480 Mpixels/s, but who cares, it's still way beyond the 366 Mpixels/s of the fastest current 3D-chips. I'd also like to note that I have yet to see a game that runs with a frame rate that would go hand in hand with those high fill rates. A game with dual-texturing would thus have to run at 95 fps at a resolution of 1600x1200, GeForce could do 125 fps at this resolution. I've never seen any results even close to that ... so far about those nice fill rates. GeForce achieves the 480 Mpixels/s with 4 parallel rendering pipelines and a core clock of (only?) 120 MHz. AGP4x Fast Write Yes! We haven't even got an AGP4x platform available on the market yet and GeForce is already faster than the rest! The 'Fast Write'-feature enables the CPU to directly write to the graphics card's frame buffer without taking a detour through system memory, and it's supposed to be up to 30% faster than 'normal' AGP4x. Let's see if Intel's 'Camino' or i820 will ever be ready and working or let's hope that 'fast write' will be supported by VIA's AGP4x 640 chipset. NVIDIA is proud to be the only 3D-chip maker that has a product supporting fast writes and NVIDIA deserves to be commended on implementing this new technology, even though I doubt that we will see much of an impact of it anytime soon. 2D as well as 3D-applications are supposed to benefit from AGP4x fast write, so you have all reason to be happy about it, regardless if you draw faster than your shadow in 3D-Westerns or type faster than your keyboard in Word.

• Total graphics market share numbers in Q4 2002 by Jon Peddie Research. Today Intel is leading the pack - thanks to its integrated chipsets. Integrated graphics are very attractive from a cost perspective, since motherboards require only minor modifications to accommodate these chips. OEMs that build low-cost PCs love these all-in-one solutions, as it saves them the extra PCB, memory and cooling solution that a discrete graphics card would need. At the same time, it is their low cost that has thus far prevented integrated solutions from being more successful. At $41, Intel charges only $5 more for its i865G with integrated graphics than for the graphics-less i865PE. At the store, an i865G board should sell for roughly $10-$15 more than the PE version. But what can you expect for $5? Computers are sold with numbers - the bigger the better. "DirectX 9 graphics" sounds better than DirectX 8, and 256 MB of video memory are obviously better than 128 MB. Many less well-informed customers rely too heavily on the bigger the better number philosophy. In its nForce 2 IGP, NVIDIA uses the NV17/18 core of its GeForce 4 MX 440, which consists of about 27 million transistors. Judging from its performance and its feature set, it's safe to assume that Intel's i865G uses far fewer transistors. The trouble is that DirectX 9 support in hardware requires very complex GPUs with a very high transistor count. Even NVIDIA's entry-level GPU, the NV34 (a.k.a. FX 5200), counts a full 47 million transistors, while its bigger siblings NV31 (FX 5600) and NV35 (FX 5900) weigh in at 80 million and 130 million, respectively. So if more power also means more transistors and consequently higher (manufacturing) cost, then integrating a low-cost GPU that at the same time incorporates features that will help sell the complete system seems impossible at this point. The issue represents an interesting paradox. But even if, by some feat of engineering or ingenuity, such a combination became possible, the chips potential would be limited by the memory bandwidth available to it. Dual channel DDR 400, for example, offers a peak bandwidth of 6,4 GB/s. While this may seem pretty respectable compared to the 8 GB/s of a GeForce 4 MX 440 add-in card, the trouble is that the integrated GPU has to share this bandwidth with the rest of the system. So an integrated graphics chip can actually be detrimental to system performance, at least when bandwidth-hungry 3D applications are used. And let's not forget that the video memory is subtracted from system memory, taking up resources. In light of these limitations, the integrated graphics solutions found on motherboards today appear to be one big compromise and seem to exist only on the fringe of the market, away from the mainstream. For our performance comparison, we assembled a group of three boards based on Intel's i865G, NVIDIA's nForce 2 IGP and SIS' 651 chipsets, respectively. Before we get to the actual comparison, we take a look at each of our contender's features and peculiarities.