NVIDIA/AMD Linux Gaming Performance For Hitman 2 On Steam Play

[humbug]

Great benchmarks. Thanks for including the R9 290 Michael, lots of us still using Hawai GPUs. It's unfortunate the development effort has stopped to move the amdgpu work past experimental.

But still fantastic to see it as a 5 year old GPU performing so close to the gtx 1060. The aftermarket models are even quicker as they maintain higher clocks without thermal throttling.

[Etherman]

Dxvk spoofs an AMD GPU by default because compability reasons (nvapi) so at least the games code path should be the same.

[betam4x]

TPU uses Intel Core i7 8700K which is not too much slower than 9900K in absolute most games.

I strongly suspect DXVK disables certain Direct3D 11 features for AMD GPUs that's why they end up running faster in Linux than in Windows.

I haven't tried hitman 2, but it is well known that DXVK can be faster on Linux than on Windows. I know this based on my own gaming with my 1080ti. Sometimes there are bugs, yes, but it's also Vulkan and a shader recompile likely happens, so there is a ton of room for optimization when doing the translation. Disclaimer: I haven't looked at the source code myself. However, I know that with proton Valve is working hard.

I am a bit perplexed by his frame time chart though. If I am reading it right, it should be inverted.

[Brisse]

This can't be right. Linux cannot run games faster than Windows in emulation (two layers of emulation actually: Win32->Posix, Direct3D11->Vulkan and not to mention that the graphics stack in Windows works close to the kernel and in X.org works as a user process in Linux).

It is not impossible, only unlikely. Imagine a case where Windows performance was crippled due to driver regressions or poor optimization and it worked perfectly well on Steam Play. The latter could win the performance race in such a case. Actually, this might be the case with HITMAN 2. Look at Radeon cards on Windows compared to Nvidia and you will probably notice how poorly Radeon performs in this game on Windows. The Steam Play results are more on par if you compare red and green team.

Also, I hate to be that prick, but Wine is not an emulator (but you already knew that I guess)

[birdie]

Keep in mind no idea what CPU was used in the Windows results you are citing where as my system was with 9900K, among other factors.

TPU uses Intel Core i7 8700K which is not too much slower than 9900K in absolute most games.

I strongly suspect DXVK disables certain Direct3D 11 features for AMD GPUs that's why they end up running faster in Linux than in Windows.

[drlamb]

The more I look at your results the weirder they become.

Vega64 at 1440p:

On Windows: 51.2FPS
On Linux: 65.1FPS

This can't be right. Linux cannot run games faster than Windows in emulation (two layers of emulation actually: Win32->Posix, Direct3D11->Vulkan and not to mention that the graphics stack in Windows works close to the kernel and in X.org works as a user process in Linux).

Or maybe it just means you can't directly compare two independent tests done on different hardware? 8700K vs 9900K.

[Michael]

The more I look at your results the weirder they become.

Vega64 at 1440p:

On Windows: 51.2FPS
On Linux: 65.1FPS

This can't be right. Linux cannot run games faster than Windows in emulation (two layers of emulation actually: Win32->Posix, Direct3D11->Vulkan and not to mention that the graphics stack in Windows works close to the kernel and in X.org works as a user process in Linux).

Keep in mind no idea what CPU was used in the Windows results you are citing where as my system was with 9900K, among other factors.

[aufkrawall]

Hitman 2 is partially not well optimized for Radeons, they lose quite a lot of performance in some foliage intense levels vs. Nvidia. On Windows, Radeons perform well in the Miami Formula 1 level as well.
But great to see that DXVK + RADV deliver top notch performance here.

Regarding the verdict which framerates are playable: Kinda disagree, just set dxgi.maxFrameLatency = 1 in dxvk.conf to get lower input latency and better frametimes at usually just a very slight fps loss. I play the old Hitman 1 levels in Hitman 2 on an RX 580 without issues in 1440p. Just the stupid screen space shadows (which are basically invisible), ultra LOD setting and reflection quality higher than low are a problem, but the game usually still looks very good with that tweaking anyway.

[birdie]

Besides DXVK itself, there's also the matter of the Vulkan drivers and the other factors involved.

The more I look at your results the weirder they become.

Vega64 at 1440p:

On Windows: 51.2FPS
On Linux: 65.1FPS

This can't be right. Linux cannot run games faster than Windows in emulation (two layers of emulation actually: Win32->Posix, Direct3D11->Vulkan and not to mention that the graphics stack in Windows works close to the kernel and in X.org works as a user process in Linux).

[Brisse]

Nice benchmarks. Thanks!

I get around 50fps (don't remember the exact number) with my Fury running 1440p medium-ish, but the benchmark is quite heavy compared to the real gameplay, so I'd say the game is perfectly enjoyable even when the benchmark shows numbers slightly below 60fps. It's not the type of game that demands insane frame-rates to be enjoyable either, and it looks good even on moderate settings.