Since I've been having a weird sort of fun estimating where Vega could end up, let's see what 1600 MHz Fury X and Polaris could do.
If the other rumors are right and Vega is:
Then let's look at Fury X, which is the same layout at 1600mhz (52% OC)
According to TPU:
Fury X vs 1080Ti: HD 60.3%, 25x14 57.6%, 4K 56.7%
Apply the 52% OC:
Vega based on 1600mhz Fury X vs 1080Ti:
HD 92%, 25x14 87.7% 4K 86.4%
Not enough to catch the 1080Ti
For Polaris let's use RX 570. It's exactly half of Fury X / rumored Vega
Standard:
RX570 vs 1080Ti: HD 47.2%, 25x14 42.0%, 4K 38.3%
Vega based on 2x RX 570 @ 1600mhz (28% OC) vs 1080Ti:
HD 121%, 25x14 108%, 4K 98.6%
So 2X OC'd Polaris is in the ball park to 20% faster than the 1080Ti.
The estimates for the higher resolutions are probably more wrong due to the 570's performance falling off sharply. Vega with 400+Gb/s of HBM2 shouldn't have that problem at 4K.
What I don't know is how much of an increase Vega will have over Polaris CU to CU.
Comparing Fury X to 2X 570 already shows an increase in performance:
RX570 x2 @1050mhz vs Fury X
HD 132%, 25x14 123%, 4K 114%
(Polaris looks to be a much better balanced design)
2X 570 is about 464mm^2 and 11.4B transistors. That's smaller than the rumored 500+mm^2 Vega. Since the 570 is a harvested 580 there's actually 72CUs instead of the rumored 64CU design. That leaves a budget of at least 1-2 billion transistors for:
- HBCC
- HBM2 mem controler (delta from the Polaris GDDR5 256bit controller)
- HPC features
- Performance enhancements
So after all this rambling I'm still left with the numbers showing Vega likely to be
at worst on par with the 1080Ti and more likely 10-20% faster. With an outside chance of it being greater than 20%.
I could be off base, but it doesn't seem likely to me.