this is alot to absorb. I would hope PS3 has a good AA method as well as enough samples of it to do away with aliasing alltogther, when it is combined with the high resolution the PS3 will have to offer.
Panajev - if cost and transistors/die-size were no object - I would advocate a super GS Visualizer - a combination of both worlds.
this device would be several billion transistors - would have both super super mega fast polygon drawing of very tiny polygons, as well as lots of hardwired based effects, as well as a powerful front end, of PUs+APUs for programmability. it would have both tremendous fillrate (512 pixel pipes!)and tremendous work being done on each pixel. in this fantasy world for GPUs, we have an L1 cache-like eDRAM of say 128MB, and these do take up transistors. but its not on a bus. its actually part of the logic. then, we have 2-4 GB of an L2 cache-like eDRAM that does NOT take up transistors, it conected to to chip on a very wide bus.
ok forget that. none of that made sense. but you can see what i was getting at. trying to solve the problem of rendering by both massive amounts of tiny polygons AND awesome hardwired effects (that are fast) PLUS the programmablity of a CPU for additional FX.
nevermind, this is all kinda worthless but I was trying
the real PS3 will probably be VERY close to your educated estimates.