Yeah, AMD truly ****ed up by showing 2x Polaris compared to 1x 1080. If we assume 34.5 FPS compared to 58 then that makes it ~60% slower than 1080 which makes it ~380X speeds. Even one of those daft slides showing up to 1.2x R9 390 or something would have been better.
If it is really 390 type performance at similar VRAM, price and performance then it will be a monumental failure with AMD as a laughing stock (again).
Didn't AMD give an official starting price of the 8Gb version as $229. I think the prices up to $300 will be premium AIB cards with higher clocks.
$229 model might sit between 390 and 390x, the $300 models with high factory clocks could be over 390x.
From the specs 390-390x performance is what is to be expected. The card if 5.5tflop, slightly less than the 390x. More importantly the TDP is half the 390x. This gives twice the performance per watt which is the most official figure that CEO Lisa Sui told investors, which has a heavy legal consequence. The 2.8x type numbers are always based on some specific theoretical circumstance. Take the 1080,MIT was advertised as up to 3x performance per watt. We know it is about 35% faster than the 980ti, has some special hardware for multi projection that increases performance by 1.7x, and does this with 25% less power. That is approximately 3x performance per watt when stacked up.
The Polaris 2.8 figure is something similar with single pass multi-projection. AMD claims that the process is providing 1.7times the performance per watt, which is exactly what TSMC and samsung are claiming, and what I reiterated on this board to the dismissal of many. This means AMD have made up 30% performance per what by architectural improvements. That is a very realistic figure for generation to generation changes, I estimate nvidia gained about 20-25% tops from maxwell to pascal.
Finally we can think back to Kyle at Hardocp's rant. It may have some Basis in reality. What if AMD were aiming for the full 40CU part and running at 1.5GHZ due to yield and process issues. This would cut the intended performance down by 15% or more and would be totally beyond AMD's control.
You got to realize that there were important reasons why AMD used GF and why Zen will be at TSMC. AMD have a financial contract in place to use a certain amount of wafers from GF each year or pay penalties. They could have prioritized their most important product Zen to go with the most advanced process, TSMC. Since AMD would have to effectively pay wafers unused at GF they get a subsidy on production there which allows dropping the price down to $200.
AMD's marketing is very focused on performance per watt, low pieces, and using xfire target high performance. They have completely avoided taking about high single card performance. 390-390x performance at 150w for $230 is the reality as far as I can see.
