- Jul 16, 2020
That’s 19%. 118/99. Which changes almost nothing.
3080: 2 to 2.14
It would need 2.23GHz on 80 CU’s and a linear increase of 2x to the rest and even still, it probably wouldn’t match a 3080. Good luck with that.
I was using the Asus numbers as it compares coolers with the same performance. I'm also not using any of the even better AIB models such as the Sapphire version, because that narrows the gap by a view more percent. But sure, the 1% difference doesn't really alter my point significantly.
I'm also not sure why you are putting 1 and 1.19 for the console performance against the 3080, as they do not take any RDNA 2 IPC gains into account. Which we know exists.
So even using the metric of 19% the 3080 is still only a hair over twice as fast as the stock blower 5700 XT. The target for desktop big Navi RDNA 2 to match the 3080 is twice the 5700 XT. So if we break this down a bit.
The 5700 XT is running 40 CUs at roughly 1900Mhz with 448 Gbps memory that is achieved over a 256 bit bus and 14Gbps GDDR6. Working it out, it is 9.7 TFLOPs.
A hypothetical 80 CU 6900 XT at 2090 Mhz (+10% frequency) with 864 Gbps memory that is achieved over a 384 bit bus and 18Gbps GDDR6. Working it out, it is 21.4 TFLOPs. This meshes with the increased clock speeds RDNA 2 is supposed to have. Historically the frequency gains for AMD have generally been around the same when moving to a better process node. RX 480 -> RX 580 -> RX 590 for example, and that was the exact same architecture.
Thus a 2.2 straight boost over the 5700 XT in raw TFLOP numbers, and and 1.92 boost in memory bandwidth.
Assuming a rather meager 6% IPC bump to be in line with Turing(and the assumed Gears 5 performance numbers), that 21.4 TFLOPs number becomes the equivalent of 22.7 RDNA 1 TFLOPs. So 2.34 times the 5700 XT in raw numbers.
Obviously I doubt RDNA would scale perfectly, but assuming only 87% scaling you end up with a card as fast as the 3080. Which is hardly surprising as the rumors and leaks have all pointed to the goal of Navi 2x to be twice as fast as the 5700 XT.
It could of course scale better or worse and that will likely determine if it's just under the 3080 as the 5700 XT was just under the 2070 Super, or over it as the 5700 XT was over the regular 2070.
If we assume (as you did) a perfect 100% scaling and 2.23Ghz (neither of which is likely), then the 80 CU card would be 2.35 times as fast as a 5700 XT or roughly 17% faster then the 3080 assuming a zero IPC increase. Add on a 6% IPC increase and you have the card being 25% faster then the 3080. But I believe this to be unlikely.
Which is too bad because that’s the point of this thread.
I originally hopped in to discuss the XSX = 2080 statement and that AMD is doomed discussion. For what its worth, I believe Ampere is going to have better RT performance, just based on what I can see from the PS5 reveal so far. That is just a wild guess though.
Which, once again, makes zero sense. If you take that figure at face value, then RDNA 2 is slower then Polaris.DF straight up said they were comparing it to GCN.
Because the blower style card sucks? Even then the difference is only 1%. The original review had drivers so bad the card couldn't even be overclocked.Why would you use benchmarks from a costlier overclocked version of the 5700xt. Seems a bit disingenuous and has nothing to do with improved drivers.