blashyrkh :
well, assuming this info is correct, the 3870 with 64 actual (we all know the story about the 320 unified shaders) is about 15-20% slower. So, going from 64 to 160 is going to give a massive boost. Also TMU count has also gone up so I'm guessing aroung 50% increase in performance translating at about 20% faster than the GTS (G92) which ain't bad at all.
If the rest of the rumours are true (ie TMU #s and ROP#s) then actually I'd say it would be even faster than that easily 50% faster than a GTS-G92 in many cases. But of course that's a guess, and baed on the way these things perform I'm not saying globally 50% faster, since even if some apps were 50% faster on the RV670 that'd simply bring it in line with the GTS-92/GTX-Ultra. But also likely compared to the G92 there shouldn't be such wildly low performance issues, even initially compared to the G92, and compared to the older generation it would not need the drivers to be as optimized to perform well, and then the driver boost would be wipped cream and a cherry on top.
Ofcource, we'll just have to wait and see.
Yep a ton of rumours and nothing concrete, right up to the G80 and R600 launches alot of the rumours seemed plausible, but still lotsa mis-fires there, even like 2 weeks before launch. The only thing I don't understand is how to rectify the transistor count and chip size for all of these added shaders. That's pretty a pretty big increase, and should involve a chip in the over billion club just like the G100/T200.
Another thing that bothers me though is the asymetrical nature of the #, I would've expected a move to 128 more than a move to 160, however the extra 32 would be welcome, but makes me question the ratios. Unlike the nV series there's no hard requirement for ROPs-Memory interface, but now it makes me wonder how they all balance. If they are only increasing the TMUs and ROPs by 2x but increasing the Shaders by 2.5X , then it's still a very shader heavy design. I just hope they not only tweak the TMU count, but also the ratios within the TMUs, now thaat would help a bit.
FInally, FSAA performance is back!! (...or it will be back...
)
Well yes, the added shaders even if not optimized and not efficient at first render should allow for near 'free AA' for current apps a current normal resolutions compared to the HD3K series. And when moving toward shader correct AA required in DX10, well that would basically be at about the same cost and would be interesting to see compared to the G92, and then also to compare to the G100/T200 if they still used hardware resolve in the ROPs.
Sounds like both future chips are gonna be monsters.