Quote Originally Posted by Dimitriman View Post
hasnt amd already said officially that adjusting the clock speeds of duzens of thousands of chips could take months not weeks to perform? why do "news" sites keep insisting that it is because it wont be competitive?
please... what a lame excuse from amd is that...
there have been half a dozen of lame excuses about why the 6900 series is delayed, one worse than the other...

cards that launch in mid december?
sounds like they will miss the holiday shopping spree...
pretty amazing that nvidia managed to get the 580 out in time tbh...

i wonder what went wrong at ati/amd...
and if the 6900 series really cant compete with gf110, then ouch... that would be a big fail from the red team...

i mean rv940 is 25% smaller than rv870 and not much slower, around 7% on average according to tpu. if your normalize the clocks (7% higher fot rv940), ati needs around 10% less transistors for the same performance with rv9xx vs rv8xx, or gets 10% more performance out of the rv9xx parts.

so an rv970 at the same die size as rv870 should be 10%+ faster, which would put it below a 480, but once you up the clocks it should be on par with a 480/570.

but rv870 wasnt very big, so it would make sense that ati felt more confident especially since 40nm is mature now, and went for a bigger chip.
i would honestly be very surprised to see ati go for the same chip size as rv870 and not more... so lets just look at how big rv970 has to be to beat a 580, and lets assume rv870 clocks, which shouldnt be a problem.

rv870 = 333M
gf110 = rv870+33% (1080 tpu av)
rv9xx@333M = rv870+10%
so we need ~23% more transistors to reach gf110 performance, on average
that would result in roughly 400M transistors... that shouldnt be a problem to make on tsmcs 40nm node these days at all...
and this is with only rv870 clocks at around 850mhz...

to significantly beat gf110 ati would need a 10%+ av lead, again at rv870 clocks that would mean ~440M transistors...
such a chip would still be ~15% smaller than gf110... and since ati knows nvidia is going for such a big chip, why would they be greedy and go for a tiny 400M or even whimpier chip? at the kind of yields tsmc has right now that would barely make a difference in cost...
and then theres scaling which isnt linear, so actually they need a bit more than 400M transistors to reach 580 performance

they have two options, go for a massive chip like nvidia, which might have issues. they would have the fastest single card, but a bit late and they would have to cut down a lot to make it a dual card, which they need to keep the ultimate perf crown.

or, they go for a medium sized chip, between rv870 and gf110, which is slightly slower than a 580 but costs 100$ less, and which combined with a second chip can easily dethrone a dual gf114 card within the 300W limit.

my guess is the latter, as its whats ati has been doing in recent years and worked well for them. what would be the point do copy nvidias strategy which hasnt worked well at all for them recently...

plus, if rv970 would be huge, we would have heard of it, im sure...
all that was leaked is that partners got disabled and fused down cards... no mention of an unusually large die size...

so my guess is ~580 performance for 6970 and ~480/570 performance for 6950, at 399$ and 299$ and at a tdp of 250W+. and thats probably whats causing ati a headache right now, heat... with higher efficiency and stuck on the same node, heat is whats limiting them, not die size...