Something about T&L
Unbalanced?
I'm not criticizing NVIDIA over this, because just putting on hardware transform and lighting is a huge accomplishment. I mean, just huge. To be able to pump out 5 times more triangles per second than the fastest Intel CPU available ... wow. But unfortunately it just isn't balanced when combined with the 480MP/s fill rate. The 3 million triangles/s by current processors isn't even being used, but the 366MTexels/s provided by the fastest 3D card out now isn't enough for our frame rate demands. Q3A at 1024x768, full detail is totally unrealistic for a competitive deathmatch. Even 640x480 full detail is too much. :)
Naturally, we should actually take Crusher.dm2 into consideration. Crusher, as most of you know, is the de facto standard benchmarking demo for worst case performance in Quake2. Even at 640x480, the limiting factor is usually the CPU, *not* the graphics card. Hypothetically speaking, if Quake II supported hardware T&L, the GF 256's GPU should easily be able to take the stress of this demo (or the even more impressive Half-Life "Smokin'" demo by the same 3Fingers who did Crusher).
Improve the worst case!
Hence, I might have to come to the same conclusion that dual processors on Q3A provide: While best-case performance will never go up dramatically, since in this case the fill rate of the video card is the limit, worst-case performance, with tonnes of calculations being done on-screen (6 player models, multiple rockets, lights, plasma blasts, etc.) is far more punishing on CPUs than fill rate, then the average framerate of the GF256 with its 480MP/s would be a lot faster than the average framerate of video card that didn't have hardware T&L but had the 480MP/s.
Thus, while I find the 480MP/s disappointing, and I'd wait for the followup to the GF256, I still like the card... even if it is unbalanced. (by followup I mean: like the TNT had a TNT2, I think we might see a
GeForce II with a faster clock speed and hence higher fill rate - most likely done on a .18 micron process, though with no extra features.)
OK, now to the 3dfx next generation part (again, I remind you, henceforth referred to as Voodoo4 or V4, for no other reason than not to look like an idiot because 3dfx thinks the super-secret secret name of
their super-secret next generation super-card has to be really super-secret. :)