Thanks for that link. So it's about 11.5% smaller, at least 1.2b less transistors, and will consume 30-40 less watts than Polaris.
I have been reading up and basing this on what i have read(i could be wrong), and gcn might be much more flexible than pascal.
I assume that pascal is more energy efficient and high clocking because the pascal architecture is less flexible than gcn, meaning everything is hardwired and optimized as much as possible already. This is why gcn improved over the years, the ace to some extent and the hws mostly from gcn (and thus polaris) can be updated with microcode. So, when (AMD or game developing or Sony or Microsoft) software engineers come up with an elegant solution, it is sometimes possible to update the gcn hws to get a performance improvement when scheduling wavefronts for the compute units.
Maybe Microsoft realized this too and Sony as well (New intermediate updates for the consoles).
For pascal, i assume Nvidia made very optimized functional units that perform top of the line and can clock extremely high. :thumbsup: But maybe can not be patched with microcode. They are perhaps hardwired, meaning they are fixed in silicon. So any issues that may arise has to be solved on the driver level. And this may also be why Nvidia cards do not age as well as when compared to AMD gcn based cards.
This is all assumption on my part. I have no proof. But it sure explains a lot.
A real comparison can only be made by software engineers that develop optimized code for both cards and know and understand the limitations of both pascal and gcn4(1.3).
I assume the reason why pascal is so much more energy efficient than gcn is that it is very possible less flexible than gcn. But of course, Nvidia is very good at predicting how long a given gpu family will be sufficient in time. When more gcn games(that favort async compute and like ace and hws) come out in the future and pascal is no longer sufficient, a new family will come out. Good economics.
See it like an asic and a cpu. An asic compression codec is much faster than a cpu. And consumes less power.
Footnote :
Also, i get the impression from reading all the reviews here and there, that polaris might have the same issues as the fury cards did, not all cu can be utilized efficiently for graphics at the moment. I think Nvidia cards get utilized better. But that will change with driver updates in time.
Also, AMD is seriously waiting for dx12 games that favor async compute a lot and make heavy use of ace and hws. Then, we will see what has been seen before in the recent past, lot of async compute will not slow down a gcn based chip much. When the updated consoles come out, (which have mostly updated ACE and HWS hardware (!). Then and only then we will see polaris shine, because then the game engines for the consoles are backported to run on the pc. And run quite well.