No announcement yet.

A Look At The Linux Graphics/Gaming Performance With GNOME 3.30 X.Org/Wayland

  • Filter
  • Time
  • Show
Clear All
new posts

  • A Look At The Linux Graphics/Gaming Performance With GNOME 3.30 X.Org/Wayland

    Phoronix: A Look At The Linux Graphics/Gaming Performance With GNOME 3.30 X.Org/Wayland

    Given last week's big GNOME 3.30 release I was eager to test the updated desktop environment with its Mutter compositor improvements on Wayland as well as seeing how its performance under the conventional X.Org Server. Here are some of these benchmarks of various graphics applications and games tested under both GNOME 3.28.3 and GNOME 3.30.0 with both Wayland and X.Org sessions.

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    Almost every game benchmark shows a slight FPS improvement under Gnome 3.30 using Wayland.


    • #3
      Pretty sure the performance improvements were on the CPU side of mutter, not something you can see in gaming, esp when in full-screen. Mutter won't do much there.
      So it would be informative to compare the CPU performance when running (windowed) apps in gnome - stuff like openoffice, gimp, etc...


      • #4
        fyi- Ubuntu 18.10 cosmic is boasting an impressive GNOME 3.30, mutter 3.30, Xorg 1.20.1, GCC 8.2.0, Linux kernel 4.18, and if you add oibaf PPA, that will give you Mesa 18.3/LLVM7 with 4.5 OpenGL compat profile. Never been a better time to go bleeding edge.


        • #5
          Well it seems like for games the big difference with the fall distributions (ubuntu 18.10 / fedora 29) will be xorg 1.20, which brought a really big improvement for xwayland. Now, apart from the clearly buggy behavior of unigine in fullscreen mode, the overhead is almost gone, although the wayland session still lacks the equivalent to unredirecting in xorg (at some point its supposed to use hardware planes).

          As for the performance gains in gnome 3.30, AFAIK they are mostly about situations where shell elements are visible. In some cases, gnome needed way more cpu power than expected and certain animations were slow. That got a lot better and overall the cpu usage got reduced in common desktop use cases. But there's also quite a lot of stuff in the pipeline which didn't made it, so with 3.30.x or 3.32 expect a lot more of good stuff to come (


          • #6
            Does Wayland still only support 60Hz refresh rate?


            • #7
              Originally posted by humbug View Post
              Does Wayland still only support 60Hz refresh rate?
              The Clutter master clock supports three synchronization methods in order of preference: (a) Hardware presentation times (b) Swap throttling (c) Fake vsync at 60.00Hz


              • #8
                Is anyone benchmarked games with SDL2 and SDL_VIDEODRIVER=wayland?


                • #9
                  Well, at least no regressions. It's still something.


                  • #10
                    Not surprising but still reassuring.

                    There was clearly a HUGE problem with mutter performance on the desktop in Gnome 3.28. I tested under Ubuntu & Archlinux. With an AMD GPU (up to date MESA etc.) just scrolling a webpage would eat 25% of the CPU. With an NVIDIA card and proprietary driver it could even reach 40% of my CPU (7th gen i3). Desktop effects were often lagging. etc. Those issues don't happen in KDE/Kwin even though it's significantly slower when using NVIDIA compared to Intel / AMD OSS drivers (in the latter case it's constantly butter smooth).

                    It's great the Gnome team finally see those performance issues as a problem. It's great more users are being affected and it's great Canonical is also providing patches. The situation is improving quickly !

                    In this area there must have been some significant progress, but how to benchmark those ? And how to benchmark Canonical's patches that didn't make their way in GS 3.30 ?