Announcement

Collapse
No announcement yet.

GNOME 3.16 On Fedora 22: Wayland vs. X.Org

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • #21
    Have you tried making the SDL2 games (e.g. Xonotic) use the wayland backend using SDL_VIDEODRIVER=wayland?
    Sadly, using the wayland backend breaks relative mouse motion in at least Xonotic (i.e. you can only turn ~300? horizontally, which obviously makes the game unplayable).

    Comment


    • #22
      Would just like to thank Michael for doing these benchmarks. Having a starting point, which these benchmarks provide, will help us as we move forward and want to verify progress. Currently there is actually more overhead with the Wayland backend because even GNOME Shell still depends on XWayland for certain things, but as we move forward we will be able to drop that requirement completely and also optimize the XWayland rendering path. So in the end our expectation is that pure Wayland applications will be faster than their Xorg counterparts (for instance GTK3 applications using the Wayland backend), and legacy apps running XWayland will be maybe 3%-4% slower (so hopefully not really noticeable) than those running on a plain Xorg server.

      Comment


      • #23
        I just did some interesting Weston, Gnome/Wayland, Gnome/X11 Benchmarks with Xonotic with the demo from here:
        http://dri.freedesktop.org/wiki/Benchmarking/

        Specs: Intel Core i7 2600K 4.5 Ghz, 8 GB RAM, GTX 580, Mesa 10.5.5, Linux 4.0.2, Xwayland 1.17.1
        Gnome: 3.16.2
        Weston/Wayland: Latest from Git
        Monitor: 2560x1440

        Frame: Min/Avg/Max
        I did 5 runs each manually and the average was very robust and only fluctuated by 1 Frame. Any difference in means greater than 3 is probably significant.

        Weston:

        Settings Low - Wayland: (Xonotic-SDL SDL_VIDEODRIVER=wayland)
        65 85 101
        Settings Low - XWayland: (Xonotic-SDL SDL_VIDEODRIVER=x11)
        43 53 58

        Gnome on Wayland:

        Settings Low - Wayland: (Xonotic-SDL SDL_VIDEODRIVER=wayland) [Shell toppanel is stuck]
        58 76 90
        Settings Low - XWayland: (Xonotic-SDL SDL_VIDEODRIVER=x11)
        45 59 67

        Gnome on X11:

        Settings Low - X11: (Xonotic-SDL)
        52 62 73

        I also tested Ultra Settings, but it's too much GPU bound and too slow to see anything significant I got 11/13/15 on all runs on Ultra.

        Conclusion:
        Gnome on Wayland is 22,5% faster than on X11 both using their native backend. Weston is 11,8% faster than Gnome/Wayland and 37% faster than X11.
        Xwayland on Weston is slower than on Gnome and both are somewhat slower than the real thing. I was a bit surprised and tested Xwayland on Weston and Gnome again rebooting after doing my 5 runs and I got the same averages for both again. Still a bit surprised how much faster it is on Weston with the nativ backend than Gnome on Wayland or X11. It's really noticable and much much smoother.

        EDIT: When I use "the-big-keybench" demo which is much longer Weston with native backend is still 26.6% faster than Gnome/X11.

        This might not be true for other games or graphics cards of course.
        Last edited by blackout23; 05-16-2015, 12:36 PM.

        Comment


        • #24
          Originally posted by blackout23 View Post
          Gnome on Wayland is 22,5% faster than on X11 both using their native backend. Weston is 11,8% faster than Gnome/Wayland and 37% faster than X11.
          Xwayland on Weston is slower than on Gnome and both are somewhat slower than the real thing.

          Given you have low end card and huge resolution, the difference you see is due to the buffer copies of the size of your screen:

          Weston (Wayland): 0 copy
          X11 DRI3 with composite bypass: 0 copy (you didn't tested)
          X11 DRI2 with composite bypass: 1 copy (what you had with gnome X11)
          X11 DRI3 without composite bypass: 1 copy
          X11 DRI2 without composite bypass: 2 copies
          XWayland DRI3 (without Present support) without composite bypass: 2 copies (gnome Xwayland)

          As to why weston XWayland has so bad perf, it is a mystery. It should be XWayland DRI3 with composite bypass, thus 1 copy.

          XWayland can get Present support (there was an implementation at some point, but wasn't updated), thus reducing the number of copies from 1.

          Comment


          • #25
            Originally posted by mannerov View Post


            Given you have low end card and huge resolution, the difference you see is due to the buffer copies of the size of your screen:

            Weston (Wayland): 0 copy
            X11 DRI3 with composite bypass: 0 copy (you didn't tested)
            X11 DRI2 with composite bypass: 1 copy (what you had with gnome X11)
            X11 DRI3 without composite bypass: 1 copy
            X11 DRI2 without composite bypass: 2 copies
            XWayland DRI3 (without Present support) without composite bypass: 2 copies (gnome Xwayland)

            As to why weston XWayland has so bad perf, it is a mystery. It should be XWayland DRI3 with composite bypass, thus 1 copy.

            XWayland can get Present support (there was an implementation at some point, but wasn't updated), thus reducing the number of copies from 1.
            Thank you for your explanation. I thought the latest nouveau also supports DRI3.

            Comment

            Working...
            X