Originally Posted by Baasha
I saw this video last night and posted it in the other thread. A few of them are saying that this guy's info is complete BS and is not to be trusted.
The numbers he mentions do seem on the high side (70% better than 2080 Ti in "some" games for instance) but meh, speculation is speculation. Also, if these numbers are anywhere near accurate, the price of the 3080Ti is not going to be cheap - may be $1500.
2080 Ti is 50% faster than 1080 Ti in certain titles and they are basically on the same node (12nm is incredibly similar to 16nm FinFet) and 2080 Ti should only be 25-30% faster in rasterization. I'm referring here to the performance difference in Red Dead Redemption 2.
The Division 2 I'm seeing a 60% difference between my 1080 Ti and 2080 Ti @ 3440x1440. Other titles also around 50% (No Man's Sky, Shadow of the Tomb Raider etc) The Witcher 3 is over 40%.
Turing and Pascal are nearly on the same node.
Ampere is dropping from 12 to 7nm.
70% uplift in rasterization and a massive increase in RT (400%?) whilst also reducing TDP to 220-230W (GA-102) is totally possible given the node drop.
We've been saying this for a while, if node drops to 7nm this will be a huge deal and this leak, while spectacular, isn't far from the previous performance estimates we've seen from CorgiKitty and Tapei Times investor call / investment meeting.
They are halving the node.
If they do halve the node to 7nm this all of this is ballpark and "Moore's Law" is correct in likening this to Pascal vs Maxwell or even Maxwell vs Kepler all over again. The 25-30% avg (with some exceptions) performance difference between Turing and Pascal was anomalous precisely because the node wasn't hardly refined or reduced in size. They simply increased the die size, made some architectural refinements, and then added Tensor and RT cores.
I'm leaning more towards this leak being factual than not. "Moore's Law" clearly wants to succeed as a tech tuber and already had a growing audience up until this leak, why would anyone in his position want to throw away their credibility with a fake leak?
I don't see motive here, and the performance cited doesn't seem to stray from what we can expect dropping the node from 12 to 7nm.