guys check digit-life.. There is an actual and (one of the best I have ever read) benchmark of the Geforce3.
And if you guys ask my opinion, GF3 starts to shine when the games listed on nvidia's site will be on the market. As for the current games, what gf3 brings is a more efficient memory access which increases frame rates in 32 bit and/or with FSAA (the new z-buffer management tehcniques and cross-bar memory switch). In their article, it is clearly seen that the fill-rate of gf3 in 32 bit is around 1100 Mtexel, which was like 600 Mtexel in GF2 ultra. It is a shame for Nvidia, because most of these techniques were already incorporated in Radeon which was released almost a year ago.
As for FSAA, if your monitor can display 1280x1024, why to stick to 640x480 with 4x FSAA. This first one will give you the same quality and saves some bandwidth due to filtering. The same applies to 800x600 as well. And in case for 1024x768, the FSAA is only required for the race games or simulations where there is a high contrast images on scene. BTW, in directx8, AA can be applied only to specific objects instead of full-screen. This means if the contrast of the textures in the environment is low (where the difference between FSAA image and non-FSAA image is barely noticable), then AA may only be applied to the portions where the other objects would be drawn, which will save a lot in terms of bandwidth without sacrificing quality. I hope the guys at Nvidia adds this as a feature in their card.
I am really curious about the sucessor of the radeon. These guys packs lots of features even in the radeon like 3 texture pixel shader (yep, this one year old card can run most of the pixel shader effects created in directx8 (and according to digit-life, sometimes faster than gf3!!!!)), volume textures, hardware shadow mapping, all kinds of bump-mapping, advanced z-buffer management techniques. And, what they could not do right in the previous chip will be completely right on the next chip.
Lastly, I hope it would not take too long to see a MX version of this core.
Best all,