Originally Posted by Slomo4shO
The only real side by side comparison we have of the two techs is the Apple A9 which utilized both fab techs due to low yields in early production.TSMC 16nm on left, Samsung 14nm on right
Toms was the only tester that got better results using samsung. Pretty much every other source out there got better using the tsmc version.
What people have to realize is that the 390x was pushed in excess of it's comfortable sweet spot for performance per watt and it's why it's performance was so good for it's die size. It was basically an overclocked 290x and it's why the difference in performance was there, even those it was the same chip with just a bit faster memory.
The same trend is even present with the gtx 1080. If we compare the 1080 to factory overclocked AIB gtx 980 ti, it actually barely wins or matches it potentially.
The thing is, Nvidia was always in a position for pascal compared to maxwell, to trade performance per watt for higher clocks. And they did this with pascal. Pascal eats into alot of it's headroom and performance per watt, to increase it's performance comfortably above a stock gtx 980 ti. Nvidia could afford to do this because the gtx 980 which the 1080 is the successor to, was a 165watt card. So it could afford to lose some of it's performance per watt, so Nvidia overclocked pascal a bit, gave it some gddr5x to save some power and turned it into a titan x beater at 180watts.
The 390x on the other hand is a 275watt card, tonga has roughly the same performance per watt as well for a 180watt card. This means AMD has alot more work to do than Nvidia. What this means is for AMD to fit 390x performance into 120 watt profile, they have to make a big sacriface as far as clocks go. This is because with both processes, there is a 60% reduction in power, when the frequency of the chip is not changed, or
a 50% increase in performance with similar wattage. The more clocks you add, the less power that is saved. So what this means is AMD is more likely to keep it's clocks similar to it's 28nm parts to translate much of it into power savings. Maybe a 10-15% increase I am thinking.
Considering the 390x is a 275watt card, AMD can't really afford to clock polaris 10 high like Nvidia and Pascal. Particularly because this is a mainstream card and many OEMS want a card that fit under 130watts and will want blower fans as well.
Add in that the a9 was originally supposed to be an all samsung order then got switched to a tsmc order as well, and there are clearly something iffy about samsungs process.