NVIDIA's Back with NV35 - GeForceFX 5900 Ultra

NVIDIA made two poor bets a couple of years back that led to the lackluster introduction of the GeForceFX 5800 (NV30).

The first mistake was one that has been widely covered - the decision to go with a 0.13-micron manufacturing process. NVIDIA assumed that TSMC's 0.13-micron process would be much more mature than it was when NV30 was ready to go to production; in the end, a 0.13-micron process that lacked maturity held the NV30 back many more months than it needed to be.

The second mistake NVIDIA made with NV30 was in their assumption that flip chip packaging technology would not be ready by the scheduled introduction of the NV30. Without a flip chip package, NVIDIA did not think that a 256-bit memory interface would be a viable option and thus chose to outfit the NV30 with a 128-bit memory interface.

Both of these mistakes combined with drivers that were still in their infancy led to a very disappointing introduction of the GeForceFX. Not being the type of company to just fade away, NVIDIA immediately shifted their focus to NV35 even before NV30 hit the shelves. The end result was everything the NV30 should have been:

Just by looking at the table above you can see that NVIDIA was able to use slower (and cheaper) memory with the NV35, as well as drop the core clock a bit in comparison to the NV30. Will the architectural enhancements and 256-bit memory bus be able to overcome these reductions? You better believe it and soon you'll find out exactly how. Also note that NVIDIA is planning on introducing cheaper 128MB versions of the NV35 core in the next month.

Today we're finally able to bring you all there is to know about NVIDIA's latest GPU and the new GeForceFX 5900 Ultra.

As if that weren't enough, today we've got a special treat for you all as well. You may have heard of a game called "Doom3" and later on in this review you'll see a total of 8 GPUs compared using a current build of the game.

Let's get to it…