Windows vs Linux gaming benchmarks comparison

[Guest]

First of all, I'm sorry but I couldn't really benchmark every game I wanted because I don't have OBS with NVENC support (and in ubuntu 18.10 is an hassle to compile)
So basically I will cover every game that runs on both Linux and Windows with "builtin" benchmarks tool

Testing platform:

CPU: AMD Ryzen 5 1600,

RAM: HyperX 2667 OC 4x4GB,

GPU: Zotac 1060 3GB Mini,

SSD: Crucial MX500 512GB,

PSU: Enermax Triathlor ECO 550W,

MOBO: MSI B350M Mortar
OS: Windows 10 (Latest)  + Ubuntu Linux 18.10 (Latest + nvidia drivers 415.23)


The game list includes the following, but I'm planning to add more games and updates in this thread. Also because I don't have many titles in my steam library...

• Half Life 2 lost coast (Native, Steam)
• F1 2015 (Native, Steam)
• GTA V (DXVK, Steam proton) 
• Bioshock Infinite (Native, and DXVK)

 

 

 

Grand theft auto 5
Let's start with this title. I used the builtin benchmarking tool which is composed by 4 pass, I made some graphs with libreoffice. The settings are the one recommended by geforce experience, FXAA=ON, DX11,very high + high + some normal settings.
Minimum frames

 

 

 

Half life 2 lost coast

Well, an old game...Obviously maxed out settings, this should make you understand I don't have many games in my library


Don't get fooled by the scale of this graph, the fps difference is just 9 fps in favour of linux, unless you got a 293Hz screen, that won't matter. Also this game is capped at 300 fps, and the results above there are the average fps. You could guess games will run just a little a bit better with this configuration on Linux.

F1 2015

 

I don't play this game often, I got it for free. But it also got a builtin benchmarking tool, here are the results



However I guess there is some sort of vsync enabled in Linux...

Bioshock Infinite, ultra settings

 

 

With "Tweaks" I mean I reduced the streaming texture pool parameter, which seems to help improving Unreal Engine performance, however, it doesn't seem to affect the results that much


Analysis

 

Let's say even if Linux isn't better for gaming in general and competitive gamers, with an high refresh-rate monitor, or  low-spec gamers, it is perfectly suitable for casuals. Fps are not that bad in very high settings, there is an initial fps drop due to shader caching for DXVK, but once it becomes big enough, there won't be any. Also, in native steam games the OpenGL/Vulkan shaders are pre-downloaded from other people using the same GPU, this is known as Steam Shader pre-caching, and also happens on OpenGL and Vulkan Windows games as well (Not sure about steam play + dxvk)

GTA 5: The first thing you probably noticed is the low minimum framerates, but there is a reason. It was the first time for me running the benchmark in this system, and as I said DXVK requires a Vulkan shader compilation right after something happens, that causes an fps drop initially. It will be stable then.
As you can see, the maximum and average framerates are perfectly fine, and playable.

Half life 2 lost coast: Well, there is an advantage for Linux, but only by 7 fps, this could be explained easily: The Linux build is a lot newer, and could include some slight performance improvements. And for well-built native OpenGL games, performance could be better than the Windows DirectX counterpart. This isn't always the case, and I will explain that later, for Valve games like Dota 2, in Linux runs much better, Is clearly not a bad linux port, also the engine could be completely rewritten for OpenGL, and not using a quick translation layer, it is just a build well-built like the "Windows" one.

F1 2015: Here Windows has a clear advantage, and shows that bad linux native ports (probably using some quick-made DirectX to OpenGL translation layer) could have worse performance than DirectX 11 games executed through a well-built API translation layer like DXVK, I remember games like saints row 4 having much worse performance than Windows too, probably for the same reason. However, for F1 2015 60 fps are still fine. Saints row 4 was running even less than 60 fps, and running the Windows build through DXVK + wine got better performance than the native linux one!

Bioshock Infinite: Interesting results, the results in Windows however fluctuate a lot on every benchmark I made, especially the max fps, also I I wasn't able to run the latest dxvk version because it has issues, at least for me, so ignore the dxvk version.
Still a decent experience on linux though. 

Conclusion


As always, DXVK is not even out of alpha, same thing applies for Steam Proton. The situation would improve in the future, couldn't say the same to the already bad Linux ports tbh, and I would suggest to give Wine with DXVK (Using lutris maybe) a try instead of bad linux ports which are using a quick and dirty DirectX to OpenGL translation without even a single improvement in the code.

 

Edit: The better explanation about Linux ports that are not truly native https://www.google.com/amp/s/amp.reddit.com/r/linux_gaming/comments/5cis3p/feral_interactives_indirectx/

 

tl;dr most game porters from Windows to Linux use a source code transition layer which makes games portable, but without the logic of OpenGL the games will have worse performance than a fully native game engine developed for a specific API

 

[Emanbaird]

This is super exciting, with windows getting more and more expensive and being a crushing blow to budget builds.

 

 This could mean you would be able to spend more on another part, (i.e Ryzen 3 -> Ryzen 5)

 

 the difference with programs like Wine used to be super big, meaning some games would run at less than 50% of  windows, if my games could run 95% the same speed of windows i would daily Linux in a heartbeat 

[Guest]

1 hour ago, Emanbaird said:

This is super exciting, with windows getting more and more expensive and being a crushing blow to budget builds.

 

 This could mean you would be able to spend more on another part, (i.e Ryzen 3 -> Ryzen 5)

 

 the difference with programs like Wine used to be super big, meaning some games would run at less than 50% of  windows, if my games could run 95% the same speed of windows i would daily Linux in a heartbeat 

Yeah but I think the overhead for things like DXVK is more consistent on non-powerful CPU's, because the transition layer between DX11->Vulkan sucks out a bit of CPU, so for mid-range builds should be okay 

[Linus_torvalds]

On 12/17/2018 at 8:02 PM, Emanbaird said:

the difference with programs like Wine used to be super big, meaning some games would run at less than 50% of  windows, if my games could run 95% the same speed of windows i would daily Linux in a heartbeat 

Well, you can always make windows games run on linux at 98% of the speed they run on windows with GPUpassthrough