by Josh Walrath

(Editors note: I did as much study as I could to delve into the different architectures of the companies involved, but I should not be considered an expert.  There could be some technical inaccuracies in this article, but it is the best information that I had at the time.  Feel free to email me if you have anything to share about this article or the information within.)

            Since the official release of the GeForce FX 5800, NVIDIA has been under attack by the press and enthusiasts everywhere for their products apparent lack of performance against ATI’s competing Radeon R3x0 series of cards and chips.  Since NVIDIA had such an extensive fan and loyalty base, many users have felt betrayed and let down by the company.  This is not to say that the users who put down $500+ US to buy the thunderous GeForce FX 5800 Ultra should not be mad about the situation, but there is a lot of background not covered by the press that should be brought to light.  There were a series of events that doomed the GeForce FX technology to play second fiddle to ATI’s R300 series in overall performance, and these events happened years ago. 

The Past

            Since the release of the NV-3 processor (Riva 128), NVIDIA has always had a performance part that performed close to the competition in terms of speed, but also included features that the others did not have.  The NV-3 was set up against the infamous Virge series, and the legendary Voodoo Graphics.  In the face of such competition, the NV-3 was one of the better selling graphics chips of its time.  It was not nearly as expensive as the dual chip Voodoo, and performance was leaps and bounds faster than the Virge series.  The NV-3 also looked a lot better and was only slightly more expensive than the Virge.

            From that point on, NVIDIA has arguably been the dominant player in the desktop graphics arena.  NVIDIA fans were rarely let down after the initial hype of the first TnT product, and NVIDIA has had a string of very successful products from that point on.  All of this changed with ATI’s introduction of the Radeon 9700 Pro.  Even after the first 9700 products hit the shelves, NVIDIA was feverishly trying to finish the NV30 design. This design was without a doubt the most complex product that NVIDIA had released to date.  Add that complexity on top of the manufacturing troubles TSMC was having with its 130 nm process, it is no wonder that this product turned out to be a disappointment for NVIDIA.

            This is not to say that the engineering of the R300 chip was far superior to the NV30, they are in fact just different.  The NV3x series of products are actually quite elegant in function, but this elegance does not translate into barn-burning performance in rendering operations.  This lack of speed due to tradeoffs on NVIDIA’s part has led to the use of questionable optimizations in drivers, plenty of positive and negative spin from press and manufacturers alike, and a general distrust and misunderstanding about what is really going on under the hood. 

Microsoft and DX9

            NVIDIA was instrumental in the development of the DX7 and DX8 spec, and their initial products always fell in line with these specs.  This changed with DX9.  When Microsoft first convened many of the graphics players throughout the world to develop the DX9 specification, one player was absent.  NVIDIA did not rejoin the talks in developing DX9 until quite some time after much of the groundwork for DX9 was laid.  Those watching the situation were quite confused as to why NVIDIA was not actively participating.  I was unable to find out exactly why NVIDIA did not show up in many of these initial meetings, but we can only assume that this was a political move that ultimately backfired on them.  Around this time NVIDIA was selling quite a few GPU’s to Microsoft for the Xbox, and the Xbox had only been out a month or two when the initial DX9 meetings took place.  There are those that speculate that NVIDIA was hoping to force its own version of DX9 on Microsoft, due to its seeming hold over Microsoft with the Xbox GPU.  NVIDIA did develop much of its own spec, and the NV3x architecture was drawn up around this.

            During this absence at the meetings, the #2 graphics chip maker worked very closely with Microsoft on developing the DX9 spec, and it already had engineering work going on with the R300 series of GPU’s.  Many of the smaller players were also present at these meetings, such as S3/VIA, Bitboyz, XGI (Trident and SiS), and Intel.  Looking at S3's upcoming Deltachrome chips and XGI’s Volari series, it appears they all supported the ATI/Microsoft DX9 standard to the hilt.  Eventually NVIDIA did rejoin the talks, but the NV3x architecture was too far ahead to go back and start from scratch.  Still, NVIDIA was able to change the initial spec into something a bit more friendly to the NV3x architecture, but it was too late to make significant changes to the overall architecture.

            NVIDIA was aiming for a much more complex specification for DX9, but the others obviously thought that the timing was not right for such an implementation.  What eventually came out in the wash was that ATI had a product that followed nearly the exact specifications for DX9 and Pixel Shader/Vertex Shader 2.0.  NVIDIA was left with a part that exceeded DX9 specs, but due to its design will never overcome the overall performance of the competing ATI part in pure PS/VS 2.0 functions.

Next: Flexibility vs. Speed

Copyright 1999-2003 PenStar Systems, LLC.