Let's follow the thread here, so we can come to some sort of conclusion.
It is demonstrable fact that the last several years of GPUs have been based upon the same manufacturing tech. 2011 is when the 28 nm node became the standard for GPUs. That was the HD 7xxx and 6xx series of cards for those counting.
It is demonstrable fact that the 20 nm node is a write-off for TSMC, and thus neither Nvidia not AMD could count on it to die shrink their cards to get a performance boost.
My sources:
http://techsoda.com/no-20nm-graphics-amd-nvidia/
http://www.extremetech.com/computin...-plans-massive-16b-fab-investment-report-says
So, what?
First off, the steps from the 680 to the 980 have been minor. If we are to follow the logic that subsequent generations move down 1 rank (680-770-960), then Nvidia has been moving along at pace. We can see that from raw numbers (given, actual gaming performance is harder to quantify, so numbers will have to do).
http://gpuboss.com/gpus/GeForce-GTX-960-vs-GeForce-GTX-680
AMD has only had one step since the introduction of the 7xxx series. You'd have to compare the 7970 to the 280. These cards see pretty much in line with a one generation slide.
http://gpuboss.com/gpus/Radeon-R9-280X-vs-Radeon-HD-7970
What about improvement? The 680-780-980 should show some reasonable differentiation. The 680 to 780 is pretty substantial.
http://gpuboss.com/gpus/GeForce-GTX-780-vs-GeForce-GTX-680
The step to 980 is less impressive. There's a great deal of compute performance, but that was a legacy of competing with the insanely popular AMD offerings for crypto-currency. Yeah, that is still a thing.
http://gpuboss.com/gpus/GeForce-GTX-980-vs-GeForce-GTX-780
AMD fares similar to the 680 to 780 leap. There's definite improvement from the 7970 to the 290x.
http://gpuboss.com/gpus/Radeon-R9-290X-vs-Radeon-HD-7970
So complaining about stagnation is reasonable, but has to be tempered by the fact that we are making progress. It isn't leaps and bounds, but it is consistent despite the lack of a die shrink for the last 4 years.
What can we actually conclude from all of this?
1) AMD and Nvidia fanboys are idiots. Neither company has fundamentally changed the market, bucked trends, or even brought us anything but incremental performance improvements since the last node shrink. Everyone moans about Intel delivering about 10% improvement per generation since Sandy bridge, but somehow allows that sort of crap from GPU manufacturers. Where are the common standards?
2) If you own anything from 2011 or later then you don't need to buy a new GPU any time in 2015. The technology is still based on the same node, and is only a minor incremental improvement over what you've got.
3) We won't see anything substantial until TSMC gets their crap together and makes the 14 nm node a reality.
4) Both AMD and Nvidia are out of juice on this process node. Nvidia demonstrated it with a poor 970 design, and AMD seems to be rehashing everything except the highest end cards. It sucks for consumers and manufacturers alike. The thing is AMD and Nvidia can't do anything about it. If they sunk huge amounts of money into R&D they might come up with a 10-15% performance improvement through sheer optimization. In 6 more months TSMC will have the 14 nm node running, and 10-15% improvements can be had simply from the shrink. Try justifying a couple million dollars in research to get an improvement the competition can compete with simply by sitting on their current product line. It can't be done, and AMD realizes that.
Bemoaning this is stupid. It is AMD admitting that 28 nm is functionally a dead end, which Nvidia did with Maxwell already. Minor improvements, that don't justify the large associated costs. Stop carrying a cross for AMD or Nvidia, and just wait for the release of something worth gushing over.
Edit:
I can't argue that AMD and Nvidia are free from error here. Both companies have decided to be dependent upon TSMC, and TSMC is basically capable of driving the bus now because they have no competition. I'd love to see some decent action between competing foundries, but that sort of investment is just too much money for any competition to develop.
On a side not, yeah Global Foundries is a twice baked turd. If I ever hear that they hit a production timeline I'd immediately check if somebody had drugged my beverage.
