f Polaris 10 comes anywhere close to the 1070, it would be a victory already.
No, it would be expected.
I consider the 1070 to be the 660 Ti of its generation. Cut down in different ways, but both are separated from the main 104 chip by a good margin.
If it is 232nm and does not nip at the 1070's heels somewhat, it is a failure.
660 Ti 294nm cutdown
7870 212nm full
At launch 660 Ti was around 10% faster.
1070 314nm cutdown
Polaris 10 232nm(?) full
Nearly the same ratio! Basically the same. Why shouldn't full P10 be within 15%? Even if the 660 Ti was cutdown a bit more overall, I'm giving a 5% margin here - and AMD should have a very slight node advantage right? Maxwell was a large jump over Kepler, but Pascal seems to just be a revision with no real IPC improvements - this is literally a rare pause in Nvidia's continual improvements. So AMD has had 4 years to make a similar efficiency jump NV made in 2 years, since Pascal apparently is mostly 2014's Maxwell on 16 finfet.
They can still price it to be a winner. But technology wise it is weak if it's not within 15% since AMD showed "New" written over almost everything with Polaris. They cannot have the same Hawaii TFLOP efficiency, or even single digit % improvements. They surely have done something more in 4 years?