the 90-80nm shifts, bought about by tsmc/tcsm/tsmc or whatever that damn plant is called, was pretty sweet in that it was so close to 90nm that it required very little work to reduce the size. but 90 to 65nm is a pretty large amount of work. it can literally mean, worst case scenario, realligning almost every ickle transistor, not likely to be that bad, but theres a huge amount of work. considering, as with basically every single company thats big at the moment, they will have multiple teams working on multiple projects. they'll have had a R700 team going for a long time. you can either lets be honest, waste time on getting a little more speed out of the 2900xt, or get going on the next card. remember, even the next gen will have been mostly done without amd's involvement i would think, these cores are worked on for a couple years.
now i'm not saying the 2900xt isn't a good card, but so far, the 2900/8800's aren't exactly looking fantastic with dx10. maybe its just drivers but, the games i've tried aren't exactly flying along with any card i have. the sooner the better for both truely new gens rather than refreshes really.
the only possibly big problem is. ok ati may have jumped the gun on the software AA, a lot of the big wigs in the gaming industry are saying IQ accurate, high IQ AA properly done with HDR now being used massively really WILL require software AA. so nvidia might be going 1teraflop with the next card, but if the game designers are asking them for software AA aswell, how much of the extra power might be used instead of hardware AA?
