By the way, if we assume a relatively optimistic 70% improvement over 290X that is still not amazing considering that by the time of the launch(likely around Computex, so early June), the 300-series would be just a few months short of two years after Hawaii was released.
So if we use a common rule of thumb of about 30% increase performance per year, that gives us 1.0*1.3*1.3=1.69, so almost 70%. That is entirely in line with expectations. Maxwell provided a 28% improvement, when you compare 780 to 980. I'm using Sweclocker's performance index, where they average a bunch of games and normalize performance. And that is really a somewhat unfair comparison since the 780 was Big Kepler while the 980 is a midrange GM204 arch.
If we get 20 nm on top of this, as well, then I'm not even sure if the 70% would be that revolutionary, then I'd say we'd need more than 100% improvement for it to really hit it out of the park, especially considering that you have HBM, too.
From a historical technology point of view, no it wouldn't be "revolutionary" since 60-70% more performance over 290X would be 2.2-2.3X faster than an HD7970Ghz but 3 years later,
a similar feat R9 290X accomplished over the HD6970.
Similarly, 7970Ghz was 75-77% faster than an HD6970 1.5 years later (same link).
You are right that on an annualized basis tracking back from HD5870/480, GPUs increase about 30-35% per year and 75-100% performance increase every 2 years or so is now unobtainable. So, no it won't be revolutionary like X800XT --> X1800XT or 5900U to 6800U.
However, I think you gotta look at the timing vs. performance increase now because we are stuck on 28nm:
R9 290X / 780Ti = Nov 2013 = 100%
GTX980 = Sept 2014 (10 months) = 115-120%
R9 390X = June - July 2014 (9-10 months) = 160-170%
Even if R9 390X is 45% faster than an R9 290X, it will already be more impressive than a 980 given the context of time.
In your 780 to 980 example, 1.5 years passed and 980 provided just
32% at 1440p, and 34% at 4K. If 390X drops with 45% over the 290X by Summer of 2015, it will have already done A LOT better than 980 did vs. 780/780Ti. It's from this point of view that R9 390X delivering 60-70% would be absolutely
mind-blowing after the lackluster gains 980 brought at $550-600 price levels. We would be back to similar gains that 7970Ghz delivered over the 6970 or 290X did over the 6970 given the context of time
I think some of us just can't believe the excitement that ensued with 970/980 cards because like you we expected 40-50% faster than 290X from a next gen flagship card. That's why to me a 980 is just a mid-range card that NV is milking now because they can and consumers keep falling for the same tactics that NV used with 680-> 780Ti transition. Other than power usage and HDMI 2.0, I am no longer impressed with the 970/980 cards as I was at launch as I got caught into the hype train and almost got dual Gigabyte 970s myself.
Also, 50-60% faster than a 290X puts us as 295X2 level of performance. That's very impressive to me from a single chip 2015 card given the slow trickle of performance since 290X/780Ti.
http://www.computerbase.de/2014-09/geforce-gtx-980-970-test-sli-nvidia/6/