Just can't believe how inefficient this thing is (power to performance wise). Is the 14nm node AMD using not as good as the 16nm node Nvidia is using or is this inefficiency due to design?
Design, Nvidia did a whole new architecture in Maxwell which don't forget still on the 28nm node was a huge performance/w improvement over Kepler. AMD has seen similar benefits from Hawaii to Polaris as Nvidia saw from Maxwell to Pascal which is primarily process. Though it's worth noting there are quite a few more hardware changes on Polaris than Pascal and some of those are certainly not being heavily utilised yet, HWS for example. But you look at Hitman, and some Doom results and intrinsics which should carry very specific performance optimisations from console to Polaris with little to no extra work.
Don't forget Polaris will be in two consoles, one in a little over a month(Xbox S One), one end of year I believe(neo) and Vega will be in Scorpio next year. So a lot of AAA titles with DX12 will have significant optimisations for GCN and more specifically for Polaris and Vega.
AMD cards are always forwards thinking, if Polaris gains a significant amount of performance using hardware that is turned on but not entirely utilised it will effectively increase it's performance/w over time. Compare hitman performance to other DX11 games and realise that almost every graphically advanced ultra demanding game from here on in will be DX12 as standard and optimised for consoles using a range of AMD hardware and in which DX12 allows the specific architecture optimisations to apply on PC as well as console.
Polaris isn't Pascal level efficient but it's a bigger jump than people realise, in 6-12 months the performance gains and consistent DX12 games will show this though. Vega will be another large step.
Effectively AMD was an architecture behind Nvidia, Polaris is AMD's Maxwell and Vega is AMD's Pascal.
People want Polaris to show a full process node performance/w improvement over Maxwell.... when AMD didn't match Maxwell in performance/w to begin with, it was an unrealistic expectation. AMD set out to make the fastest ~$200 card, which they've done, it's selling extremely well for them. By the time Nvidia have their top to bottom Pascal range out, AMD will have their Vega and Polaris range out. lack of a new architecture at the end of 28nm hurt them, but this is moving them past it in one generation. Navi will be targeting Volta up and down the range as most generations.
