Something very interesting i found while testing my 2 different graphic cards (on i7 + 32gigs) :
GTX 770 (4gigs of DRAM)
R9 290X (4gigs of DRAM). I bought it recently because it's reported as outperforming GTX770 by 30% (average).
So i did a test using High settings for benchmark with DX11 & DX12 (1920x1080).
And guess what ?
With DX11 GTX770 absolutly outperforms R9 290X !!!!
With DX12,
GTX770 gets a slight better result than with DX11, but basically same result.
R9 290X outperforms GTX770 as expected, and yes, DX12 results are 2x better than with DX11.
A very interesting observation with DX11 is that DX11 cpu overhead reported for R9 290X is twice higher than for GTX 770.
So one may guess what's the point of this post ... While DX12 seems to be the best possible direction and i really thank Oxyde for having pushed Microsoft to move forward on optimizing the DX API thanks to Mantke, I can't prevent myself from thinking that AMD is really doing a poor stuff at optimizing DX11 drivers.
And while it seems obvious that Nvidia should consider a new architecture optimized for DX12, The current one is absolutly running as expected with very good dx11 drivers.
Since i do not have any other Nvidia card (like GTX980) i can't figure out if Nvidia architecture is limiting DX12 results, but for sure, on DX11 AOTS is likely to run really better on Nvidia than on AMD if AMD does not provide better drivers. It really looks like to me AMD is pushing out DX11. But what about people who will not move to Win 10 ?