A new benchmark video shows Dota 2 with Vulkan performing better on Windows than Linux

Quoting: liamdawe

Quoting: devland

QuoteWhat are your thoughts, any theories?

Vulkan is a new API. So far all performance tests yielded similar results. This was expected and is not news.

Please don't sensationalize the issue. Vulkan will get better in time as will it's implementation by game developers. :)

No one is sensationalising anything. These are the facts.

Quoting: rkfgFrom reddit:

QuoteMaybe the Linux driver is still not perfect for the 1080 card?
From Phoronix, discussing the latest beta driver:
The NVIDIA 367.18 Linux driver doesn't explicitly mention any Pascal support for the new GTX 1070 and GTX 1080, but I believe that's in there too. The next driver release though should officially mention this new graphics card support.
So that suggests to me that the tested driver (not sure of the Linux version he used, I assume the latest stable) is not great for the 1080.

Liam, you have Windows and you can test the renderer yourself. 980Ti should be supported very well and this concern would be struck out.

I don't see the point, since I have a 980ti, not the 1080 that was tested. Totally different generation.

I actually see a point here. Your card is powerful and supported AFAIK. It means that you can do the benchmark of DOTA2 just to compare your results with his (I'm not interested in all modes he used, just the best ones from Windows (including Vulkan) and Linux Vulkan/OpenGL. I would be interested to see such a comparison.

Quoting: reaVerI'm still calling for this to be tested on a high performance distro rather than Ubuntu.

What, exactly, is a "high performance distro"? Every benchmark I've seen suggests that the performance differences between mature distros is negligible, especially when it comes to gaming.

----------

Quoting: MblackwellTo the above question about borderless window: No, if an application is not exclusive fullscreen it won't set the flag required to bypass the compositor.

That's a great point. Running with compositor enabled is a great way to kill gaming performance on most Linux desktops.

Quoting: melkemindI'm actually more surprised Nvidia's Linux drivers already suppirt their 1080 card.

Driver was out in Arch main yesterday. Nvidia 367.27-1.

It's in the changelog.

I think it is worth asking they guy if he can check wether using real full-screen makes any difference, but this is a suspect. With borderless window, there is a ccompositor that can decrease performance, with native full-screen this should not happen. If there is still a significant difference with native full-screen, then we need to look deeper, such as into drivers.

To the people asking about the fullscreen mode used, it was stated in the article:

QuoteI spoke to the person who did it on twitter, he mentioned both Windows/Linux were done in borderless fullscreen mode.

Yes, which is not the optimal mode to benchmark. We need to know if it did influence the results.

Quoting: liamdaweTo the people asking about the fullscreen mode used, it was stated in the article:

QuoteI spoke to the person who did it on twitter, he mentioned both Windows/Linux were done in borderless fullscreen mode.

Maybe borderless windowed?

I don't understand, there's more fullscreen than borderless fullscreen?

Well, the options in Dota are Exclusive Fullscreen, Windowed and Borderless Window.

Whenever the term "borderless" is used, it indicates windowed. Borderless fullscreen just makes no sense. There's no fullscreen with borders. But a windowed can of course be maximised to fullscreen. I think that's what we talk about here.


Last edited by Beamboom on 15 June 2016 at 4:04 pm UTC

Quoting: liamdaweTo the people asking about the fullscreen mode used, it was stated in the article:

QuoteI spoke to the person who did it on twitter, he mentioned both Windows/Linux were done in borderless fullscreen mode.

It sounds like he's confused, because "borderless fullscreen" isn't a thing. There's fullscreen, and there's borderless window which can occupy the fullscreen, but just plain "borderless fullscreen"? No such beast exists.

The fact that he was running in a borderless window presumably with the compositor crunching away in the background makes me feel better about these results. Someone needs to do another benchmark using actual fullscreen mode or with the compositor disabled at the very least.


Last edited by Mountain Man on 15 June 2016 at 4:11 pm UTC

Quoting: dubigrasuWell, the options in Dota are Exclusive Fullscreen, Windowed and Borderless Window.

Not trying to impose anything, but is it possible for you to do a benchmark for Dota2 Vulkan? You are my favorite source when it comes to Linux/Windows benchmarks. You are doing an awesome job. :)

http://steamcommunity.com/groups/steamuniverse/discussions/1/541907867770822642/

It might not be the problem, but I feel, too, that using a brand new GTX 1070-1080 may not give a fair portrait of Linux performances. Usually it requires a couple of driver's iterations to get it's full capacity.

Sigh. This again..

Unless someone with Linux as a primary system who knows how to use the Linux Desktop properly then benchmarks something im never taking these as literal.

Why do they need it as a primary ? Well for one there is less confirmation Bias ( yes i believe Linux users tell things the way they are more in general than a windows centric press, they have nothing to hide and nothing to lose ). Also what compositor is used ? what settings on the compositor ? what desktop environment ? were multiple tried? what power settings / config settings on the driver ? Was there something other or weird about the testing process ? What kernal/Distribution ? How long has the card been out given that windows gets support for new hardware first ?

Im not denying that X11 is complete pants in 2016 (we have wayland with AMDGPU-PRO) and that windows does have a faster pipeline ( that's what billions of $ and two decades of dominance does ) , but the fact remains unless the above is met it cannot be a fair performance comparison.


@Liam it would be nice to get some testing between the two from some people on here who have both OS. Or in some performance articles because if we always take our benchmarks from Phoronix or the MSM gaming press there probably never going to be right.


Last edited by on 15 June 2016 at 4:40 pm UTC

Quoting: Mohandevir

Quoting: dubigrasuWell, the options in Dota are Exclusive Fullscreen, Windowed and Borderless Window.

Not trying to impose anything, but is it possible for you to do a benchmark for Dota2 Vulkan? You are my favorite source when it comes to Linux/Windows benchmarks. You are doing an awesome job. :)

http://steamcommunity.com/groups/steamuniverse/discussions/1/541907867770822642/

It might not be the problem, but I feel, too, that using a brand new GTX 1070-1080 may not give a fair portrait of Linux performances. Usually it requires a couple of driver's iterations to get it's full capacity.

Quoting: Mohandevir

Quoting: dubigrasuWell, the options in Dota are Exclusive Fullscreen, Windowed and Borderless Window.

Not trying to impose anything, but is it possible for you to do a benchmark for Dota2 Vulkan? You are my favorite source when it comes to Linux/Windows benchmarks. You are doing an awesome job. :)

http://steamcommunity.com/groups/steamuniverse/discussions/1/541907867770822642/

It might not be the problem, but I feel, too, that using a brand new GTX 1070-1080 may not give a fair portrait of Linux performances. Usually it requires a couple of driver's iterations to get it's full capacity.

I'm doing one now for my own curiosity, but not so extensive (with 4K and so on), just 1080. I might add a graph for it if you're curious, but video no.
Don't have much time and I think I'll wait a while for things to settle down (with Dota itself & drivers) before doing a video.

Another Comparision

It`s all about optimiziation... still opengl can get the same results like directx...

you should give a shit on a video from someone in the internet... it`s just one result... it says nothing...

Quoting: dubigrasuI'm doing one now for my own curiosity, but not so extensive (with 4K and so on), just 1080. I might add a graph for it if you're curious, but video no.
Don't have much time and I think I'll wait a while for things to settle down (with Dota itself & drivers) before doing a video.

Great! I'd be happy if you could share us your graph. Thanks!

Quoting: mr-egg@Liam it would be nice to get some testing between the two from some people on here who have both OS. Or in some performance articles because if we always take our benchmarks from Phoronix or the MSM gaming press there probably never going to be right.

As much as I would like to, it is very time consuming especially when I myself don't know the first thing about the Dota 2 command line benchmark tool.

If someone can PM me with details on how to do it easily, with a really heavy scene, then I can look at running a test on my 980ti on OpenGL, Vulkan across Windows and Linux myself.

...Besides...

Slice of life:

I'm working for a CAD company and we just upgraded our workstations to Win10... Not new workstations. Just a full and clean Win10 reinstall...

Conclusion:
if the situation apllies to gaming too, I can live with lower fps if it permits to avoid loads of BSODs... Crap! I've never seen that much BSODs in my entire life! Not one workstation... On all worksations with diffrent hardware configurations ranging from i7-2600 to i7-4790 with quadro gpus. Yuk! Yuk! Yuk!

End of slice of life, but still funny. :)

Just did the test with his demo file on:

Distribution: Manjaro KDE
Driver version: linux44-nvidia 1:364.19-6
Kernel: 4.4.13-1-MANJARO
Graphics card: GTX 760
CPU: Intel(R) Core(TM) i7-2600 CPU @ 3.40GHz
Resolution: 1920x1080,
Composit modes: disabled & enabled
Window mode: exclusive fullscreen & fullscreen borderless window
Graphics mode: vukan&opengl
Number of runs: 10 each graphics mode & OS

Result:
- No difference between exclusive fullscreen and borderless window fullscreen. They seem to set the X flag there.
- No difference between composite enabled or disabled (would hit it that they set the flag, kwin is checking for it to suspend composite)

Vulkan: 58-117
OpenGL: 38-115

OpenGl stuttered from time to time, Vulkan was completely fluent.

================
Result on Windows10, driver 368.39:
DX: 83-113
Vulkan: 71-122


So I can back his data, looks pretty similar here too, though, I have a lot weaker graphics card and bet my CPU is pretty much less up-to-date than his ;-).

I usually play with vsync, so in my particular case: I couldn't care less in dota2, but there is work to do if our driver should be as good as the Windows for other games reaching the same performance.


Last edited by STiAT on 15 June 2016 at 6:04 pm UTC

It seems only fair to run the tests on SteamOS in my opinion. Seriously Ubuntu is trash