One of the problems that both Nvidia and AMD face as fabless companies is the seemingly lack of availability of 40nm wafers at TSMC. You would have to imagine companies like Apple/ARM get the lions share of the capacity and AMD and Nvidia have to ration what's left with between there different product mixes (Enthusiast, high end, performance, budget markets etc), you only have to look at lack of available cards to see that TSMC holds all the aces in this battle as they control how much a company can effectively produce.
Thats almost completely untrue, TSMC for YEARS has almost completely relied on AMD/Nvidia to fill up capacity on the new process, theres no lack of wafers at 40nm, theres a lack of demand for not particularly big increasesin 2nd gen 40nm products. 40nm sucked balls, most of the people who would go 5870>6970 would have done so near release because we're talking ultra enthusiasts, most other people aren't upgrading for 20% performance. Upgrading for 70-80% performance increase on a new gen, lots of people would still be buying, Dell and co know AMD are only months away from doubling performance for LESS power usage, with smaller cores and cheaper to them, so Dell and all the other OEM's aren't particularly interested in 6xxx/580/570/560 parts either, and frankly 580/570gtx parts have never sold well and again most people that would want them being high end enthusiasts who are more likely to buy near release than near the release of the next gen.
28nm is going to be a BIG increase again exactly because its on a new process and a process that by all accounts isn't as buggy as TSMC 40nm(which for all intents and purposes was the worst "new" process around in the past decade).
Very very very few companies go for ultra high end new processes when they are brand new and expensive because unlike £300+ gpu's, most other things made are MUCH cheaper cpu wise. An Apple Arm based ipad cpu is like $30, not $200 like a gpu, its not particularly well suited to new processes due to cost, capacity and so on. I think its been mentioned before, I did some rough calculations, 15k wafers is about where 28nm will open, thats WAY more than enough for AMD/Nvidia to start off with and theres normally only 1-2 very small customers for a brand new process node, TSMC currently produce something in the region of 14million wafers a year across all nodes and for the high end which tends to launch first, AMD/Nvidia don't need a huge amount of wafers in the first place. That 15k will pretty much increase month on month till they are making them at a rate of a millions a year.
Lejosh, you said 45 to 32nm, none of the HD7000 series is that, then you said, IGP's from 45nm to 28nm, theres none of those either.
You said on die gpu's in a CPU, there weren't any of these at 45nm, full stop, at all, anywhere, it just didn't happen. Llano/Sandy with proper APU's both launched on 32nm, neither will move to 28nm.
The ONLY gpu's going from 40 to 28nm are the gpu's in a bobcat based ultra low end gpu, you're talking 80 shaders, not for real gaming, no real power, and they aren't coming out for months on end, nothing to do with whats being talked about here.
28nm HD7000 gpu's are the next gen discrete cards of which the last generation was made at 40nm and were also discrete cards.
