How on earth do you get that if the 6970 was built on 32nm the 7970/680 would have been faster, they wouldn't have. 32nm wouldn't make 28nm transistors smaller. For a 384bit bus in the 7970's case, with GCN 1.0 architecture, there would be a set amount of rops/shaders that would suit a 384bit bus and thus we would have always ended up with a 350mm^2 part. For Nvidia they went with a 256bit bus and there is a set amount of shaders that works well with that.
No but 32nm would have made 40nm transistors smaller DO not forget Cayman was scalled down from it's original design as it failed to use the half node shrink. Now imagine it came out with the rumoured 1920 shaders and not the 1536, ok gcn vs vliw4 are totally different but then look does a 2048 gcn core part over a potential 1920 look high end, even 1536 vs 2048 despite gcn efficiency does not look high end, it looks 50% gain and mid rnage.
One thing we do know is Cayman should have been faster than it was. Why would that have made the 7970 and 680 faster, well effectively the 6970 would have performed faster than it did meaning that the 7970 would have had to provide more performance than it did when released. As the 6970 was underperforming the gap between 7970 and 6970 was large, but the gap between the 7970 and the 580 wasn't so great. the 7970 could have included one more cu on each engine making it a 17x64 x2 = 2176 shaders or they could have just released the clocks at their potential of 1050-1100 which they can all seemingly manage.
yields on the Gk110(notice GK100 got cancelled because their early attempt was as rubbish as GF100) were dire. It was sold to the professional market first because it took time to build up an inventory of parts for a consumer level market, they cost/yield wasn't going to cut it for profit at early stages and releasing a full part didn't happen for a VERY long time from launch.
YES already been said in the past and recently.
Nvidia moved up their midrange part at an early date because for three generations they had problems with their 500+mm^2 parts. Midrange meant 256bit bus and however many shaders they would have calculated would work with that bus. Put 3000 shaders on it and they'd be a huge power draw and have no where near the required bandwidth to be fully utilised, put a 384bit bus and increase die size and power significantly for too few shaders and you get an inefficient card that is under powered, add the shaders to the new bus and you had a GK110 which they couldn't make.
32nm had precisely no impact at all, in any way, on what AMD/Nvidia made at 28nm.
It did on the basis of it allowed Nvidia and Amd to market mid range offerings at a higher price point and higher tier.
When a range of cores have the same architecture and are all being made at the same time, IE all GK chips, they aren't reacting to the opposition, they are just doing what manufacturing allows. Sometimes you choose to go midrange first, sometimes high end. For years Nvidia went high end, midrange 6 or so months later and low end often up to 18 months later(or not at all... see the GT230........).
With AMD the 7970 was the high end part, 290x is a different architecture, it wasn't in the making at the time the 7970 launched(GK110 was absolutely in the making when the 680gtx launched), it has a bunch of different features. They aren't related in any way and the time gap between them pretty much confirms that the 7970 wasn't a "bumped up from midrange" part as Rroff seems to think.
The 680gtx very much was one, because anyone even half awake knew about the GK110 6+ months before the 680gtx launched and knew a bigger faster core was coming when they could make it.
.
