NV40 Unveiling

By Josh Walrath

            It was no secret that April 13, 2004 would be the day that NVIDIA launches its latest and greatest graphics processor.  Sure enough, the invitations were sent out and reservations made.  San Francisco here I come.  NVIDIA was doing a few new and interesting things with the event itself, but I will go into that a little further on.  The big question that stood out in everyone’s mind was if NVIDIA would release a truly remarkable GPU, or would they stumble again as they had with the NV30 architecture?

            If we go back some 1.5 years, ATI released a revolutionary product in the form of the Radeon 9700 Pro.  In response NVIDIA tried frantically to get a solid, competing product out within the same time frame.  For many reasons the NV30 was very late to market, and even though the hype around it was still at a fever pitch, it wasn’t until samples hit the market was it found to be second best to the Radeon 9700 Pro.  This was very apparent in PS 2.0 applications, and the NV30 (GeForce FX 5800) just could not keep up with its slower clocked rival.  Some claim that only around 100,000 NV30 were ever produced during this time, and this was probably a good thing.  Right around the corner, in Spring of 2003, NVIDIA released a refresh part that would double the floating point pixel shading power, and the NV35 was released.  Again, ATI trumped this release with the Radeon 9800 Pro, which again outclassed the NV35 in terms of overall performance, especially in PS 2.0 applications.

            This is not to say that the NV35 architecture was inherently bad, it just wasn’t all that great at PS 2.0 operations.  In DX7 and DX8 apps, as well as OpenGL, the NV35 was as fast if not faster than the competing ATI product.  Still, it was second best in the world at what it did, and that very fact cost NVIDIA a fair bit of marketshare and mindshare.  NVIDIA did its best to keep the NV3x architecture alive and competitive, and to do this quite a bit of work was done behind the scenes.  First off NVIDIA had to get a solid compiler into its drivers to more adequately feed the NV3x pipeline, so rendering performance could be optimized and increased.  The second thing NVIDIA did was do total shader replacements for popular applications.  Some of these shaders had identical output to the included shaders, while some other replacement shaders did things slightly different.  This of course led many users and readers to cry foul on NVIDIA’s practices.  While I cannot condone NVIDIA’s action, I can neither decry them for trying to increase their product’s performance and usability for a large portion of customers.  While the output will probably differ from the reference image, the differences were usually not that great and the increase in performance truly made up for it.  This of course is my opinion, and others will disagree with it.  If we lived in a perfect world, then we wouldn’t have need of such things, but of course we don’t live in a perfect world.

            At the outset the NV40 has been billed as a “no holds barred” architecture that will address all of the mistakes that have been made in the past, and set up a solid base for future products.

The Fan Base

            NVIDIA has been known widely for its brand loyalty, with many users purchasing generation after generation of NVIDIA products, and with the buyout of 3dfx the brand loyalty at NVIDIA seemingly increased by an order magnitude.  Even during the rough times of the GeForce FX 5800/Ultra series of cards, these products were in short supply around the world because so many enthusiasts bought them up while they could!  The fact that there were people willing to lay down upwards of $500 for a video card that did not perform as well as ATI’s Radeon 9700 Pro (which was at a significantly lower price) shows exactly how loyal that fan base was.

            At what appears to be a nod to this fan base, NVIDIA did something quite unique with its release of the NV40.  Not only was NVIDIA going to hold a presentation for the general media, but they were also going to hold a large BYOC LAN at the same time, and invite those gamers to attend the official release.  500 spots at the LAN were opened up, and those spots quickly filled up.  I think it is great that NVIDIA opened up the release to its rabid customer base, as going to these presentations is usually a lot of fun (not to mention you can be one of the first to learn about these exciting new products).

Next: The Fans and The Show