Originally Posted by Schmuckley
The 2080S is ~22% faster than a 5700XT on average per TPU, and per their latest review it is 12-19% faster depending on resolution.
This GPU has 25% higher bandwidth and 23% higher raw performance than a 5700XT not including any IPC gains from RDNA2 or tricks such as bare metal coding or having RT hardware.
Why is math hard? And why are you offended? This is a good sign for the next generation of GPUs.
Originally Posted by Liranan
The hell is this? 10GB @ 560 GB/s, 6GB @ 336 GB/s
Are they doing a 970 on buyers?
Can't be, they're not lying about it.
But no, the 970 had 1/8th of it's RAM double stacked on a single controller segment which could only be accessed solo. Microsoft has chosen to use mismatched dies (6x 2GB, 4x 1GB) and you simply can not access the last 6GB of RAM as quickly because that "part" of the RAM is only 60% as wide as the rest. It's like running dual channel with a 4GB and 8GB stick.
The picture is actually misleading, because you can still access data in those last 6GB at the same time as you can access 4GB in the other dies. I imagine it will be a slight headache to the developers, but for all I know they just treat each die separately and don't stripe data.
The extra RAM could also be CPU/OS specific, and thus not need the full bandwidth.