Would such an increase in performance (
70%!!) over the 290x be attainable on 28nm?
Considering AMD has competed sucessfully time over time with nvidia using smaller dies at least since the RV670, the strongest example of that would be the RV770 (256mm²) / GT200 (576mm²) era... Remember that slide that showed up a few months ago, stating work done on chips sized >350mm² (which turned out to be Tonga IIRC) and one >500mm² on 28nm?
This one. The company is supposed to be AMD from what I remember. Let's suppose that chip is what we're seeing here. Let's suppose AMD decides to go nvidia (> 500mm²) on die size. Haven't we all ever given that a thought before?
Hawaii is 438mm², I can't see how much bigger this would have to become to get 2*290x kind of performance, save for a significant improvement in architecture (beyond what GCN 1.2 did for Tonga) or something along these lines... in addition to HBM. GDDR5 did the trick for RV770/4870 after all... GM200 seems to be ~630mm², but AMD wouldn't even think of going that big. Nvidia neither... up to now.
70% is insane. It reminds me of the earlier generations that relied on smaller nodes... Either there's some black magic going on, or the 20nm rumour
might (and that's as far fetched as it gets) have some grounds... or what's more likely, the image is 'shopped. I'd really like to believe it isn't, the graphics card market has become quite boring lately.
Hope 390x/GM200 bring this kind of performance to the table, 4k on a single card would be awesome.