The Geforce4 is very similar in architecture to the Geforce3 except for the vertex power and clock speed. Do you really think that its a driver bug? Nvidia had a year to release a card that was hardly different, just faster. Don't you think they would have got all the bugs out?
I really think that they had to make some compromises in the hardware. The Geforce4 has 63 million transistors, and the Geforce3 has 57 million. The Geforce4 has double the per-clock vertex power, an improved memory controller, slightly improved pixel shader capabilities (ps 1.2 and 1.3, I believe, were not properly supported by the Geforce3?), and no performance loss for Quincunx (that useless AA blur filter
).
The big thing, however, is the vertex processing improvement. How would NVidia double the power with only 6 million more transistors? Unless the Geforce3 was very transistor-inefficient, they must have cut out some abilities. When the Geforce3 came out, NVidia was fairly proud of their anisotropic filtering. I think they saw how good the Radeon and Radeon 8500 performance was, and how very few sites paid attention to it. If you look at reviews, I'd guess 70% look at AA, but very few look at anisotropic filtering (maybe 10%?). Thus they made a strategical move to strip out some of this hardware, and as far as a marketing move, I think it was a smart one.
However, I think anisotropic filtering has a greater impact on picture quality, so I would not applaud such a move. However this is just a theory.
In short, I don't see how NVidia could make a mistake in their driver if anisotropic filtering was the same as in the Geforce3. I really think it's a hardware problem.
