No one really knows. But what I can say with a high degree of confidence is, big Vega will be better than 1070/1080 which is what 99% of the people here predicted/expecting it will be. So as long as AMD get pricing right, there will be a lot of happy/surprised people at least. Few upset/angry ones too. Lol
Agree with this. Even using the flawed Fiji arch, a 500mm^2 die (seen at Vega Preview) @ ~1000MHz will have 1080 performance. Incorporate the huge changes to Vega arch, in terms of efficiency and higher clock frequencies, and it is a good bet that Vega will greatly exceed 1080 performance.
I just can't fathom that they could have lower efficiency arch than Fiji:
Fiji has 8.9B transistors and a 596mm^2 die - 15M transistors/mm^2
Polaris has 5.7B transistors and 232mm^2 die - 24.5M transistors/mm^2
The (Very) rough maths goes, that Vega will have 12.25B transistors. That's a 37% uplift in transistors alone. Then add Raja has been quite transparent of AMD trying to achieve higher clocks with Vega.
On a different line of thought, if you did a linear scaling of Polaris (Rx480) performance per transistor, it would net a 114% increase in perf before any arch improvements. I gather 1080 is 50-80% quicker than RX480 (user bench) for reference.
Obviously I'm assuming all transistors go towards performance, which is absolutely not the case. But it gives some weight to the 500mm^2 die variant of Vega >1080 claim.
What I love about this is that the 1080 has 7.7B transistors! Significantly less than Fury :/ SO much room for improvement!
Edit: The worry is if AMD can only beat the 1080 and not the 1080ti with a 500mm^2 die. Very costly!