If you make the assumption that the NV25 is approximately the spring refresh of the NV20, you could guess some of the features the NV20 will have this fall.
The most important seems to be the hidden surface occlusion, saving both texture read and zbuffer memory bandwidth. And it seems like we may see a huge jump in triangle throughput, too. Possibly 50-80 million vertices/sec?
And it seems like they've squeezed some memory bandwidth savings when using FSAA, too. Somehow also related to hidden surface occlusion and zbuffer. Looks promising that's for sure. At least they seem to be addressing the memory bandwidth issue.
BP6, Celermine 566@850, 320meg, CL TNT2U 175/220
[This message has been edited by Freon (edited 07-05-2000).]
BP6, P3-700@770 (for now), 320mb, eVGA MX Plus (dual CRT), waiting for VIVO module