Announcement

Collapse
No announcement yet.

The Cost Of Running Compiz

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

  • #16
    The scientific approach

    It happend so many times before but this time I really had to register and write this post. Phoronix should really pay more attention to numbers and their math. For example, in the "Unigine Heaven v2.0 1980x1080" we see that the Metacity result is 74% faster than Compiz, not 63%. I haven't found a rounding of the numbers that would actually lead to 63%. If Metacity leads to 25.34 fps and Compiz to 14.56, then Metacity is (25.34-14.56)/14.56=0.74... times faster than Compiz. On the contrary, Compiz reaches just (25.34-14.56)/25.34=0.425=42.5% of the performance of Metacity. Now question to the reader: By what percentage is Compiz slower than Metacity?

    Overall, Phoronix should definitly come in contact with someone with a more academic background. I'm happy that the Phoronix Benchmark Suite incorporated variance measurements to help with the significance. But this is definitely not enough. The test scenarios for the article should be designed more according to the "scientific approach" to do evaluation. Otherwise, Phoronix just targets readers who are just able to understand black/white messages. "Scientific depth" is the keyphrase here.

    Comment


    • #17
      There's several imo unexpected things in this benchmark considering radeon drivers (other parts of the test were mostly uninteresting).
      1) It seems that Compiz according to these benchmarks has a worse effect on DRI2 than DRI1 when it should be the other way around. Why?
      2) One test shows DRI2 having better performance with Compiz than without. Why?
      Some developer could possibly enlighten us with an explanation attempt?

      Comment


      • #18
        Retest 10.04 against Win 7?

        Nice Tests,

        a new comparison between Win / MacOS and Lin without Compiz/Aero/??? would be interesting. (hoping that my favourite OS could be more competative)

        In my opinion there should be a workaround in compiz (or in Gnome/KDE), that automatically stops/restarts desktop-effects on "blacklisted" applications.
        This also would circumvent problems like Vsync issues, that some drivers have.
        Of course it would be better to fix the driver issues, but I dont really believe in it...
        Has anyone a better solution than writing a shell-script for each game?

        Comment


        • #19
          Originally posted by nightmorph View Post
          What good does it do to push 400 FPS when your monitor only refreshes itself at 60hz?

          Even the low FPS (in the 30s to 60s) reported by the open-source drivers should be sufficient.

          The article is useless because the benchmarks ignore the monitor issue -- in practice, you'll never see the different between 100 and 400 FPS. Your screen simply does not refresh itself that many times per second. Most screens run at 50 to 70 Hz at max resolution.

          I find it offensive that more people don't understand this. In a game, if you're running at 60fps, each frame is taking 16ms. If you're catching input for the next frame while drawing the current, that's an input lag of 33ms between what shows up on the screen. If you're running at 120fps with the same update/display logic, the input lag becomes 16ms, a noticeable difference to gamers who actually game or musicians, among others. The fact that you'll basically drop the first frame b/c of your monitor's limitations doesn't mean it's not worth catching input, sending it to the network, etc. When you count network updates, you can see why people care (hopefully). There's a lot more going on than display. Games really need to run as fast as they can. Some games are smarter and run the update loop 2x times more than the renderer, but I'm not sure about quake 1-3 based stuff.

          Comment


          • #20
            what they do on consoles is lock the frame rate, and slow down the input anyway, so everyone's on the same level playing field, but that's silly for PC's.

            Comment


            • #21
              Originally posted by dl.zerocool View Post
              Hello,
              Sad to see of performance angry compiz is.

              I really would like to see a test with kwin.
              And a comparative test between gnome and kde for playing games.
              The should also report the problems encounter in each platform.

              Best regards.
              The most meaningful comparison between KDE vs Gnome (Kubuntu vs Ubuntu in this case) will be with sound enabled, because Ubuntu is using PulseAudio and this consumes some of the CPU power, so Kubuntu should be faster in games. Kwin should disable compositions, but it seems it's not always working.

              @Sacha

              So where does this 40-50% performance drop come from in Ubuntu? I'm guessing you can't blame Ubuntu.
              Nice catch. So it looks not only compiz is slowing Ubuntu down in games. Btw. treating Ubuntu as a 'primary' Linux OS is starting to be a shame imho. Not only they released with serious issues (which still aren't fixed), but it has also slower graphic. Personally, I can't wait for Fedora 13 final release which is looking to be the best Fedora release ever - beta is rock stable and very fast (KDE version).

              @Eoxx

              I know that KDE/Kubuntu users are a minority
              In your dreams. ;>

              Comment


              • #22
                I only play one game and that is UT2004. I have always needed to turn of Compiz when playing as I am a bit finicky when it comes to smooth gaming. Manually disable Compiz is not very convenient, so I used a script for that.

                Code:
                #!/bin/bash
                
                kwin --replace &
                ut2004 ;
                killall -s USR2 compiz.real
                compiz --replace & 
                exit 0
                This will stop Compiz when you start UT2004 and start it again when you quit UT2004 automatically.

                This script is for KDE so you would need to change "kwin" to "metacity" (and "UT2004" for a different game) if you use Gnome. It has been a long time since I used this script (used it when I used KDE3) so it might not work any longer.



                Now I use KDE4 and I made this script:
                Code:
                #!/bin/bash
                
                kwriteconfig --file kwinrc --group Compositing --key Enabled false
                kwin --replace &
                ut2004 $@ ;
                kwriteconfig --file kwinrc --group Compositing --key Enabled true 
                kwin --replace
                exit 0
                This script will probably only work for KDE4. Perhaps it would help somebody.

                Comment


                • #23
                  Originally posted by mum1989 View Post
                  Good test but...
                  I don't understand why you don't turn off Compiz when you play games
                  Because not everyone still has a computer from the stoneage...

                  I'd rather have a moderate graphics setting and a repsonsiveness and hassle free desktop environment. I am also not going to turn of 3G on my phone whenever I make a call to preserve battery life. How's a desktop computer any different?

                  PS: ow wait sarcasm

                  Comment


                  • #24
                    The behavior of fglrx is very interesting here. I see two possibilities for it:
                    - it's very much optimized for composited environments, texture_from_pixmap and all that
                    - it does equally bad under both, because there's about equivalent unnecessary operation in the direct case comparable to compositing overhead

                    First wouldn't make sense, as no enterprise distro runs compositing.
                    Second would be just plain bad.

                    Bridgman, can you shed some light?

                    Comment


                    • #25
                      Originally posted by kraftman View Post
                      Nice catch. So it looks not only compiz is slowing Ubuntu down in games. Btw. treating Ubuntu as a 'primary' Linux OS is starting to be a shame imho. Not only they released with serious issues (which still aren't fixed), but it has also slower graphic. Personally, I can't wait for Fedora 13 final release which is looking to be the best Fedora release ever - beta is rock stable and very fast (KDE version)
                      The combination Ubuntu and NVidia sucks. It has been proved before. That's why I asked the crossplatform benchmarks to include and another Linux distro besides Ubuntu.

                      Comment


                      • #26
                        Does it even make sense to include DRI1-based open source drivers in the benchmark? I think it doesn't!
                        DRI1 doesn't support compositing and simply draws directly to the screen, no matter what.

                        Comment


                        • #27
                          The "tests" are useless since they only look at stupid pointless games and other worthlessness.

                          I suspect that it won't have ANY measurable effect on any measurements of anything that is actually ***USEFUL***. By USEFUL, I mean things that REAL PEOPLE ACTUALLY DO, like WEB BROWSING and WORD PROCESSING.

                          Comment


                          • #28
                            Originally posted by Eazy View Post
                            Code:
                            #!/bin/bash
                            
                            kwriteconfig --file kwinrc --group Compositing --key Enabled false
                            kwin --replace &
                            ut2004 $@ ;
                            kwriteconfig --file kwinrc --group Compositing --key Enabled true 
                            kwin --replace
                            exit 0
                            This script will probably only work for KDE4. Perhaps it would help somebody.
                            Uhm how about Shift+Alt+F12?

                            Comment


                            • #29
                              Originally posted by brent View Post
                              Does it even make sense to include DRI1-based open source drivers in the benchmark? I think it doesn't!
                              DRI1 doesn't support compositing and simply draws directly to the screen, no matter what.
                              Yes it does because according to benchmark compositing is faster in DRI1 than DRI2.

                              Comment


                              • #30
                                Compiz makes my system a hell of a lot faster and more responsive but then, I don't do much gaming. It's downright annoying when I have to turn compositing off, or have to use unaccelerated compositing ala Xfwm4.

                                Comment

                                Working...
                                X