Anti Aliasing + more on fill rate

Nick Triantos of Nvidia stated that "Full scene AA in any system will cost performance. If you don't believe that, you're being misled by some marketing folks. :-)" Can you comment on that in regards to T-Buffer technology?
There is no doubt that full-scene AA is going to cost you fill-rate performance. We have never stated anything to the contrary. What we have said, however, is that with full-scene AA enabled that there is still sufficient fillrate in our next-generation products to sustain 60 fps at 1024x768 resolution with 32bpp pixel depth for many games. It makes sense that Nick would say something like that when he's got to promote a competitive part that doesn't support any anti-aliasing, supports no new T-Buffer cinematic effects, and has such limited fillrate capabilites.

We believe full-scene AA is going to absolutely reset the image quality standard for next generation 3D accelerators. Once people see their favorite games rendered at real-time frame rates completely anti-aliased, they are never going to be able to look at aliased rendering again (regardless of how many triangles there are in a scene!). Trust us, we've been staring at anti-aliasing algorithms for quite some time now and it's miserable looking at aliased renderings now that we've become so accustomed to the amazing visual quality improvements associated with anti-aliased rendering.

And the biggest problem with the GeForce in this regard is that they are selling a future promise that games will eventually take advantage of their geometry acceleration, albeit at a substandard fillrate. The full-scene AA capability of our next generation products upgrades every single game currently available. Consumers won't have to wait 9-12 months for new games to take advantage of new graphics technology. The day a user plugs a T-Buffer enabled graphics accelerator in their system, their library of games is immediately upgraded with substantial visual quality improvements as a result of the ground breaking anti-aliasing technology.

Nvidia has been VERY tight-lipped about the texel fill rate for their card, although they openly claim 480 million pixels/sec at what we assume is 120MHz. FiringSquad is assuming it does 1 texture per pipeline for 480 million texels/sec. Nick Triantos just stated in an IRC chat that GeForce256 does "Quad texturing...Sorry...It has a four-pixel pipeline, and can render 2 textures." What is your take on it in light of its quad-pixel pipeline (Nvidia's site claims "Four independent pixel-rendering pipelines deliver up to 480 million 8-sample fully filtered pixels)?
It appears as if the raster engine in the GeForce 256 is really just 2 TNT2 raster engines put in parallel (with some feature additions added). So, I believe you're correct in that each pipeline handles a single texture, so when you're rendering a single-textured triangle you'll get 4 pixels-per-clock rendering performance. And when you render a dual-textured triangle you'll get 2 pixels-per-clock rendering performance. The GeForce does not support more than 2 textures per triangle in hardware.

One improvement in the GeForce rendering architecture is that there is no performance penalty (at least in theory, hard to say if this is true until GeForce boards are available) for trilinear filtering. The reason nvidia says "8-sample fully filtered" instead of "trilinear filtering" is that Real3D is claiming to own intellectual property rights to trilinear filtering, and nvidia is trying not to get into trouble with this claimed patent. What lovely legal games we are forced to play in this crazy market!