Originally Posted by dagget3450
This benchmark seems to be exhibiting the same issues i get in DX11 on my Furyx quadfire. First there is still some sort of cpu overhead issue in quadfire furyx that was in dx11(guessing). I am seeing light or bouncing loads on gpus. The same thing is required like in DX11 i have to use a faster cpu(newer) to try to help lessen the load issues. Also, during the test to confirm this i also have similar power usage numbers at the wall as well. It's not using all gpus to the max constantly. The third thing that seems to corroborate the issue is gpu temps are too low as if they aren't really using much power.
So my next question which is probably a resounding no... does this bench not use DX12 multi adapter instead of CF/SLI as well?
Have awaited for someone to notice this.
Multi-GPU on DX12 is still... "needs some work in drivers".
Problem is, with full screen mode, the GPUs end up starved of work. 2 GPUs scales fairly well, 3+ starts to show this.
If you (under custom) run in Borderless Fullscreen, you'll notice interesting things...
Sadly our idea of running the whole test in Borderless Window by default was vetoed due to issues related to performance effects of Windows compositing on different resolution displays (ie. on some vendors, if you have 1080p monitor vs. 4K monitor places considerably different load when Windows composites everything in Borderless Fullscreen)
So for now, 3+ GPU multi-GPU is held back by the way drivers handle this in LDA & exclusive full screen DX12.