"and some operations of the NV2A will be up to 8x faster than the NV20."
Certainly doesn't sound like 8X faster then NV20 to me, just it will do some operations up to 8x faster. This is fairly easy to achieve if say the NV2A supports a feature in hardware that the NV20 doesn't. Big difference then saying that it is eight times faster.
The GF2 can do some operations up to six times faster then a GF1, that is real. The GF2 certainly isn't six times faster then the GF1 though.
Dave-
"You say you had a demo that pushed the GF DDR to 5 million vertices/sec. However, NVIDIA claims 15 million. This demo was designed to show of T&L too. So that means in best case you are getting 5 million, so real world I'd guess you'd get about 2 million. This means you are getting about 1/7.5 of what NVIDIA is actually claiming. So if it is being claimed that NV2a is 8x faster.. how fast is it really?"
I'm pushing over twelve million using GLPerf(AGP 2X), not a nVidia specific/optimized test, on a DDR. Running the MS DX8 "Optimized Mesh" I'm hitting over 5 million vertices/sec running AGP 1X(haven't tried AGP2X yet).
I think 15 million would require perfect optimizations, but I don't doubt that they can hit it with a GeForce if I can hit twelve million running GLPerf which certainly isn't tweaked for nVidia's T&L. In game? I don't know, as soon as developers start pushing heavy poly games maybe we will have a better idea.
[]i]"The real question is.. what will be, what is better: X-box or GameCube."[/i]
In terms of graphics, I think that much of this will come down to the artists/modelers working on the games. We are at the point now where artistic talent can start to be used in 3D games without having to make enormous compromises, much like 2D games on the PSX/Saturn. My money is on Nintendo, mainly because of Miyamoto, to be the overall better system, but I think that they both will clearly be significantly better then Sony's offering(and I plan on picking both up).