This is to be expected when comparing D3D11 to the native Vulkan implementation in a game that is notoriously slow with D3D11.
This is to be expected when comparing D3D11 to the native Vulkan implementation in a game that is notoriously slow with D3D11.
Right, I should have worded it better. DX11 performance is good on Windows, but DXVK performance has frequent dips into the low 28 FPS.
According to the benchmark, Normal & Medium batches are within 10FPS of each other compared to Windows vs DXVK,, but when it comes to heavy batches there is a ~20 FPS difference which is significant at 4k resolution.
Even with Linux set to the lowest settings, it can’t match the FPS on Windows (crazy settings) on Long Shot 3 in the benchmark. All other locations in the benchmark show higher FPS then Windows.
(left, Linux/low settings. Right, Windows/crazy settings)

Yes, same resolution. The screenshot is smaller on the right because it's
an image being displayed on my secondary monitor.
On Fri, Mar 15, 2019 at 8:58 PM Robin Kertels [email protected]
wrote:
Just to get that out of the way, are you sure you ran both at the same
resolution?—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
https://github.com/doitsujin/dxvk/issues/969#issuecomment-473496571, or mute
the thread
https://github.com/notifications/unsubscribe-auth/AHkWiKqbmFFFU64Jew2T2MatLoJukGkXks5vXGv3gaJpZM4b0IE0
.
Lower performance than Windows is also expected, especially when compared to Nvidia's Windows drivers which are significantly more CPU efficient. I don't know what kind of magic they are doing to achieve this, but you shouldn't expect the combination of Wine + DXVK + Vulkan driver to compete with that.
I honestly don't really care about this specifc edge case, the "Heavy batches" part of the benchmark is essentially a synthetic test that generates way more draw calls than any other game out there (you can show the numbers with DXVK_HUD=drawcalls), in practice it's good enough to outperform at least AMD's D3D11 driver in many games in the CPU-bound case and run them just fine. There's no obvious bottleneck that cuts performance to half of what it should be.
Ashes of singularity, "Crazy" settings @ 4k resolution, 2xAA. Game runs with esync disabled.
I know the game can be played with native Vulkan, but while running DXVK the overall average FPS is good. However, it frequently dips in the 40's to low 25's during the GPU benchmark. When running Vulkan, it's usually in the 60's with one or two dips to the 49-50's.
I don't know if this can be reproduced on a lower resolution. As you can see in the screenshot, during the FPS drops GPU utilization is also down by around 40%. It's normally at 100%.
It's possible this isn't a DXVK issue at all and a problem with how DX11 is implemented in the game, but throwing this out there anyway.