10-Way NVIDIA GeForce Linux Vulkan Performance For F1 2017

[prazola]

Why would you say there's no abstraction layer?

I'm just saying that it's one or the other and that I don't think vulkan would outperform DX11 on nvidia cards, even without an abstraction layer.

[frosth]

I think other games support RadV too. But this is the first Vulkan title which supports only RadV and not amdgpu pro. You could see the tide turning for a while though with Valve heavily contributing to it, and feral switching focus. Once it's open sourced and once it works on modern distros and kernels the tide will turn for AMD's official driver. For now Game Devs are not gonna support drivers which only workstation users run. Even steamOS switched to the open drivers for AMD.

Like with opengl (radeonsi) feral's games work better with radeonsi but you always see "this driver is not supported". There is difference between "oh it works on radv" and "radv is supported by developer".
I agree with you about reason they support radv than amd-pro-vulkan.

[bug77]

If there isn't an abstraction layer in use the problem resides in both GPU architecture and driver. Even in windows Dota 2 with vulkan requires an higher GPU usage to reach the same DX11 fps with lower minimum fps. In windows nvidia cards have higher overheads also in dx12 vs dx11. The only think they can optimize is the software part of the scheduler but I think they've already squeezed out everything.

Why would you say there's no abstraction layer?

[prazola]

GOL compared windows vs Linux and sadly it's looking really bad. On a 980ti Linux Vulkan performance is 30-40% lower than windows DX11
https://www.gamingonlinux.com/articl...t-report.10652

Ouch

Would be very interesting to get a Feral developer technical comment on the reason for the wide gap. Michael?
Would be a lot less work than bench-marking 34 GPUs

If there isn't an abstraction layer in use the problem resides in both GPU architecture and driver. Even in windows Dota 2 with vulkan requires an higher GPU usage to reach the same DX11 fps with lower minimum fps. In windows nvidia cards have higher overheads also in dx12 vs dx11. The only think they can optimize is the software part of the scheduler but I think they've already squeezed out everything.

[liamdawe]

Gaming on Linux used the 387.22 driver.

From Feral:

As noted in our article, we tested on both driver series and the performance difference was basically nothing. The main issue with the later driver is game crashes, not performance (from what I saw and from what Feral told me).

[Michael]

I vote for Windows tests. Why should it be time consuming? I mean, except for writing a Windows-compatible profile obviously. Shouldn't be as easy as plugging your external SSD with Windows and fire PTS?

Because the writing a Windows-compatible test profile will likely take a few hours, not to mention testing for ensuring similar settings between OS, etc.

[darkbasic]

Only if there were many premium supports requesting it... Time consuming benchmark and would need many changes to even get this test running on Windows in a comparable manner.

I vote for Windows tests. Why should it be time consuming? I mean, except for writing a Windows-compatible profile obviously. Shouldn't be as easy as plugging your external SSD with Windows and fire PTS?

[chimpy]

Usually feral games get better performance over time, so the performance for this game at launch is actually pretty good.

[Michael]

It would be interesting to see a comparison with 387.22 because it has "* Improved performance of fullscreen Vulkan applications using X11 swapchains."

387 currently has regressions with this game, which is why it wasn't used.

[md1032]

It would be interesting to see a comparison with 387.22 because it has "* Improved performance of fullscreen Vulkan applications using X11 swapchains."