Announcement

Collapse
No announcement yet.

NVIDIA XRender Performance Improved

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

  • NVIDIA XRender Performance Improved

    Phoronix: NVIDIA XRender Performance Improved

    Towards the middle of last month, NVIDIA had released the 169.04 Beta Linux Driver. The change-log was quite lengthy and what we had discovered while benchmarking the GeForce 8 series was that there were improvements to be found in this release and it was far more than a simple version bump. One of the reported changes for this driver release was "improved RENDER performance", and out of requests from readers and interest by the Linux desktop community at large, we have conducted XRender benchmarks using render_bench and have the NVIDIA results available today.

    http://www.phoronix.com/vr.php?view=11522

  • #2
    Try GtkPerf also!

    Btw, 1000 iterations in the "Text"-test is a bit wierd, it turns into a CPU test instead.

    And here's an older test with lots of configurations that shows Render helping a bit at least: http://gtkperf.sourceforge.net/index...e=testing&id=2

    Comment


    • #3
      And this is good in what way?

      Comment


      • #4
        Originally posted by conholster View Post
        And this is good in what way?
        I truly hope you are kidding!

        Is there any chance of seeing some older cards compared similarly? I'd be curious to see what a low end 6200 and a mid range 7x card can do. I pick these because they are the onboard graphics for the Nvidia chipsets...

        Also how does this apply to real world usage? Will I see a performance increase in in only specific applications or will everything GL benefit?

        Comment


        • #5
          The result sounds really good.

          I hope NVidia would add the hardware H.264 decoding into their Linux driver in the near future. I still have some movies that I have to reboot into Windows in order to play.

          Comment


          • #6
            I came across the following benchmarks as well:
            http://www.nvnews.net/vbulletin/atta...7&d=1196719662

            what's missing from the phoronix results (and the above) is a comparisons with the frglx.

            Also kudos to the radeonhd guys, since until recently they were better than nvidia's binary blob.

            Comment


            • #7
              Originally posted by psyopper View Post
              I truly hope you are kidding!

              ...

              Also how does this apply to real world usage? Will I see a performance increase in in only specific applications or will everything GL benefit?
              I think you asked the very same question... and i'm also interested by the answer !

              Comment


              • #8
                its totally nuts how much of an improvement this is, nearly unbelievable i wonder what they changed :P

                Comment


                • #9
                  oh well, i would like to see a large test when this new driver is stable. a comparison between the nvidia, nv and nouveau-driver would be really interesting...

                  Comment


                  • #10
                    Originally posted by Regenwald View Post
                    oh well, i would like to see a large test when this new driver is stable. a comparison between the nvidia, nv and nouveau-driver would be really interesting...
                    How'd you read my mind?
                    Michael Larabel
                    http://www.michaellarabel.com/

                    Comment


                    • #11
                      Benchmarks are benchmarks, should there be any improvment in real usage? I've had the driver installed since it was released, but I dont think I've honestly noticed any diffrence...

                      Comment


                      • #12
                        I've done a bunch of benchmarks on the new beta driver and xrender, and render_bench is the only benchmark that shows these great results without any tweaking. Specifically, setting InitialPixmapPlacement to 2, at least on my 8800 gts.

                        A better benchmark of xrender, that shows results closer to what you would get from real applications, is expedite. And yes, evas software rendering is still almost twice as fast as xrender in the overall expedite results.

                        expedite
                        InitialPixmapPlacement=2
                        -----------------------------------
                        nvidia-169.04 xrender - http://pastebin.ca/790277
                        nvidia-100.14 xrender - http://pastebin.ca/790281
                        nvidia-169.04 X11 - http://pastebin.ca/790279

                        xrenderbenchmark
                        -------------------------------------------
                        100.14 - http://pastebin.ca/784454
                        169.04 - http://pastebin.ca/784457

                        xrenderbenchmark
                        InitialPixmapPlacement=2
                        -----------------------------------
                        100.14 - http://pastebin.ca/790287
                        169.04 - http://pastebin.ca/790286

                        Comment


                        • #13
                          Yeah !!! good news !

                          It will be a very very intersting feature for e17 users... because Evas can render by two way, software, or Xrender.

                          Great news!

                          Comment


                          • #14
                            Soooo... if Qt is compiled with XRender support then KDE should be more responsive in 2D, right? What about Gtk and Gnome?

                            Comment


                            • #15
                              Originally posted by platform View Post
                              Soooo... if Qt is compiled with XRender support then KDE should be more responsive in 2D, right? What about Gtk and Gnome?
                              From what i've seen, the only way to get these results is to set InitialPixmapPlacement to 2, which has the side effect of slowing down almost all gtk apps by a very significant amount. I'm not sure about qt apps, but I would guess it's the same.

                              Comment

                              Working...
                              X