Windows 10 Creators Update vs. Ubuntu 17.04 Linux Radeon Gaming Performance

[dungeon]

The ddx doesn't matter. The issue is that some older apps use the old gamma API which stopped working when randr 1.2 support was added.

Well i didn't checked really, but remembering Michel's comment in a bug how it matters... here:

https://bugs.freedesktop.org/show_bug.cgi?id=27222#c26

[agd5f]

Better ever than never, that bug was there... something like 6+ years i think

And yeah maybe to mention that radeon/amdgpu ddx's are required too, cos maybe some people use modesetting which does not handle that

The ddx doesn't matter. The issue is that some older apps use the old gamma API which stopped working when randr 1.2 support was added.

[DrYak]

I think most of the difference comes down to game-specific driver tweaks in catalyst

I would also suggest the reverse is true too:
DirectX-specific tweaks in the game code. Most of the cited game have been initially designed with DX11 in mind and had the shit optimized out of it.
OpenGL was usually either an after though or done through some compatibility layer which add to the performance degradation.
Same with Vulkan : few games currently available were from the ground up optimized for Vulkan. (For the obvious reason that Vulkan didn't even exist yet, back when the games were designed).
Vulkan support instead was only recently added and isn't as well optimized, showing lower perfs even if in theory it should behave much better.

You can see it even on windows on the Talos benchmark :
- The game was initially designed for DX11 and it's the most mature back-end it has. This back-end dominates the banchmark.
- OpenGL even on Windows was added after it, and isn't all that mature yet. Even on windows (i.e.: on the same driver as the above DX11) it runs slower.
But the Linux version actually runs in the same range as the Windows OpenGL version.
- The Vulkan back-end is obviously recent (of course given the recent birth of this library). As a result, it's currently running slower, even if in theory vulkan/DX12/metal -style libraries should have much better performance than older DX11/OpenGL (specially non-AZDO).
But still, Vulkan on Windows already running better than its OpenGL counter-part (though not yet on Linux, which is to be expect, RADV being very early in the development and opensource drivers from AMD being still not available yet).

As the opposite you can check the openarena benchmark, or refer to past Metro results (or internal Valve tests) :
- This games were developed for OpenGL to begin with (openarena is a distant descendant of quake3 engine, the first primary optimized for OpenGL. and since then the tradition has been kept, openarena is as much optimized as possible for OpenGL)
- The Linux version runs actually better than the Windows version.

This actually gives us a few very good news :
- As one would guess, there's nothing inherently wrong with Linux. It runs quite fast actually. (and in some case provably better. See Valve's own internal testing if you don't trust the openarena benchmark).
- Probably a huge reason of the lower perfs is due to less optimized ports (see Talos/DX11 vs all other libraries on all platforms)
- That means that as more cross platform games are released (because, comeon, look at the current gaming landscape : PCs with Windows are just "yet another machine" on the long list of each relased title), the gap is going to progressively close.
- Vulkan (eventually to be supported by nearly anything with a modern GPU under the sun) is going to increase the trend of well-optimized multiplatform code (because again, makes things simpler when your release target 10 differnt platforms. And also is supposed to have better perfs than OpenGL), so long-term Linux will end-up being less disadvantaged than the rest.
- RADV is quite an amazing result for such a recent drivers (which is partly due to Vulkan being by design a much thinner layer above the hardware, and partly due to the devs being quite some genuises).

[Okki]

agd5f : do you think the driver can still be really improved concerning video game performance? Is it a goal for the developers paid by AMD, or are they only focusing on cash-generating areas, such as cloud computing and others GPU compute related works?

[theriddick]

Overall not bad results for the R9 Fury, the RX580 appears to have some issues here and there however so not sure whats going on there. Once AMD's vulkan driver joins the open scene I think that will be a game changer for quite a few things.

[dungeon]

That's actually an xserver bug rather than a driver bug and it's been fixed in a recent xserver.

Better ever than never, that bug was there... something like 6+ years i think

And yeah maybe to mention that radeon/amdgpu ddx's are required too, cos maybe some people use modesetting which does not handle that

[humbug]

You can tell exactly which games are using some sort of compatibility layer, and which ones are native. It's good to see Linux beating Windows when they are both running the game natively.

most games implement Linux as an afterthought.

the source2 engine on dota2 gets good Linux performance because it was built to be multiplatform. CS:GO will also get the engine within next couple of months

[TheOne]

If I recall correctly metro redux titles performed better under linux than windows because they are good opengl games.

[AJenbo]

Nice article

[eigenlambda]

better than half performance everywhere. This means that, in a 500$ PC build, the money saved from using Linux instead of Windows can go to upgrading the CPU from a Pentium 4560 to a FX-6300 and the GPU from a 4GB RX470 to an 8GB RX480. Or maybe keep the Pentium and get a 6GB 1060.

This means that when games become more broadly available on Linux, Microsoft is going to have to slash the price of Windows for system builders.