Here's my little take:
The 7900GT and GTX cards will out-perform anything ATI has, especially in SLI. We've seen X-fire to be a pretty substancial mis-fire, while SLI has evolved quite nicely.
G80 should ship out a month or two before Vista, there's no point in shipping it earlier when there's no DX10 out, and no competition. Also, if it's on a 65nm process, it will take a little longer than 2 months to get the bugs worked out when they're just now getting 90nm production up to par.
The big question is when will nVidia and ATI start looking at alternatives like the CPU guys have for performance per watt. Those X1900XTX cards are sucking more power than the rest of the whole machine, and with Mis-fire and SLI, video cards easily eat more juice than two systems. I'd like to see a move towards more pipelines and lower frequencies, with the emphasis on making cooler, less hungry cards that can still outperform the current models. You won't need two actual chips, or two dies, for that to happen; GPU's are already very scaled. We just need to take the parallelism another step further.
The 7900GT and GTX cards will out-perform anything ATI has, especially in SLI. We've seen X-fire to be a pretty substancial mis-fire, while SLI has evolved quite nicely.
G80 should ship out a month or two before Vista, there's no point in shipping it earlier when there's no DX10 out, and no competition. Also, if it's on a 65nm process, it will take a little longer than 2 months to get the bugs worked out when they're just now getting 90nm production up to par.
The big question is when will nVidia and ATI start looking at alternatives like the CPU guys have for performance per watt. Those X1900XTX cards are sucking more power than the rest of the whole machine, and with Mis-fire and SLI, video cards easily eat more juice than two systems. I'd like to see a move towards more pipelines and lower frequencies, with the emphasis on making cooler, less hungry cards that can still outperform the current models. You won't need two actual chips, or two dies, for that to happen; GPU's are already very scaled. We just need to take the parallelism another step further.