So the fact that Nvidia already has their own cloud gaming infrastructure has already been brought up. Just to flesh that out a bit more, I'm looking at the
FAQ page for GeForce NOW.
According to that, the different tiers have access to the equivalent of the 1060, 1080, 2080, and 3080.
The 1060 is 10 SMs on TSMC 16FF; 4.4 billion transistors and die size of 200 mm^2.
The 1080 is 20 SMs on TSMC 16FF; 7.2 billion transistors and die size of 314 mm^2.
The 2080 is 46 SMs on TSMC 12FF (which is refined 16FF for better density); 13.6 billion transistors and die size of 545 mm^2.
The 3080 is 68 SMs on Samsung 8; it's a cut down of a 628.4 mm^2 die with 28.3 billion transistors.
Now they did say the
equivalent of, so it's not gonna be exactly those cards. But the general idea should remain the same: if selecting for cloud infrastructure usage, your thinking of when to be concerned about yields is going to be very different than compared to say, when selecting for what to be used in a Switch-type device or even something like the XSS.