Same here: 165W under Linux, 240W under Windows, but same frequency.
I also notice the temperatures, higher in Windows, which is consistent with a higher Wattage.
I get way less FPS in Black Myth Wukong on Linux than Windows on a 4090
Windows 11
Linux (CachyOS)
nvidia-bug-report.log.gz (2.0 MB)
I deleted my window partition awhile and haven’t reinstall it since, I remember the game used to drop in frequency alot a couple of months ago in previous seasons, but that no longer the case (this might be related to nvidia’s reflex). GPU usage is also different DirectX12 is only pushing 82%-87% in the screenshot, compare to 99% on DirectX11. When playing The Finals on Directx11 my gpu usage never dropped 98% in game at any point in time.
If I do remember correctly I did try running the game on DirectX11 normally on Windows 10 using the -dx11 command I did notice stuttering (like DirectX11 on linux) and lost only 5% of my FPS vs DirectX12 on windows.
I edited the post to be able to download the log and have the image on the nvidia servers and not discord this time.
nvidia-bug-report.log.gz (2.0 MB)
you literally got 50% less performance on Linux than on Windows, that’s alot
That’s with ray tracing on, which is another issue in itself. Raster performance usually is 20-30% less than Windows, but with RT enabled it could be much more, like this.
should we open a separate issue about ray tracing performance on Linux?
I think the two issues could be related, but heavy RT amplifies the problem, possibly because it is CPU heavy. RT performance in Indiana Jones, for example, is practically the same on Linux and Windows, so maybe it’s not RT, but RT in DX12 games.
Yeah, RT in vkd3d makes it even worse. However, let’s focus on raster performance here.
Slightly OT, but AMD got about 20% loss in RT too… but that’s with RADV, it’s better with amdvlk I think.
I some how got more fps on linux than windows with low settings in Black Myth Wukong
Linux
Windows
nvidia-bug-report.log.gz (1.6 MB)
I think if you run the Benchmark on low it hits the CPU harder than the GPU. So youre comparing CPU performance. But you still see the way higher maximum FPS on Windows.
more examples of degration of performance using vulkan beta drivers 570.123.06
Game: Manor Lords
-
DirectX11 FPS 147
-
DirectX12 FPS 108
nvidia-bug-report.log.gz (1.7 MB)
I also noticed that with a 5080, Baldur’s Gate 3 runs on 4K ultra at 150 fps in dx11, but only 58 in Vulkan mode? I read that BG3 Vulkan implementation is not great, and that it could simply be dx12 based, therefore requiring the same kind of translation calls than vkd3d. Is this issue related to the dx12 performance issue?
EDIT: it seems that the Vulkan implementation in BG3 is indeed very bad, so not really a good benchmark for perf. IJ Great Circle and RDR2 have indeed fantastic Vulkan performance.
Which game is that?
Forgot to add title name, the game is Manor lords
Thanks. I tried to test dx12 vs Vulkan with Red Dead Redemption 2, but only Vulkan works (and works well). DX12 crashes at launch. It is good to have several game candidates that can use both APIs to compare. So far, there is The Finals and Manor Lords then. I saw also that Strange Brigade was the first game to support both dx12 and Vulkan, i should check that out.
I just tested Strange Brigade, 4K Ultra with my 5080 using latest 570.133 drivers, Proton experimental and gamemoderun + mangohud. Interestingly, using their benchmark, DX12/VKD3D has better performance than Vulkan: 213 fps and 176fps, respectively. Not that this has any impact on playability in both cases - the engine for Strange Brigade being very ancient. But this shows that the DX12 performance loss is seen in more recent engines, possibly exacerbated with RT (but not only) and possibly by some bug in the handling of more recent dx12/Vulkan features that have appeared after Strange Brigade.
EDIT: this is a game from Rebellion. Hence for a more recent engine I could check out Sniper Elite 5 by the same editor. Apparently it also supports both dx12 and Vulkan.
OK I just tested Sniper Elite 5 from Rebellion, same settings: 4K Ultra with my 5080 using latest 570.133 drivers, Proton experimental and gamemoderun + mangohud.
And as I suspected in my previous post, now Vulkan runs better than DX12! I have not found a benchmark in SE5, so I just opened a saved game and compared FPS when just looking around in the 3rd mission, keeping the same place. Vulkan was around 120 fps, whereas DX12 was around 90 fps.
In addition, despite this high fps count under dx12, it seemed less smooth. It is as if frames are dropped or something, but I do not want to overreach. So between Strange Brigade and Sniper Elite 5, Rebellion has added new graphical elements that impact the performance with VKD3D/DX12. I do not know how this is helpful, and if this is the same bug we see responsible for the lack of dx12 performance.
PS: I have no Windows partition, so I cannot ascertain there is any difference in Windows too. This would be useful to do the same test under Windows to confirm this is a Linux driver problem.





