Announcement

Collapse
No announcement yet.

videos benchmarks on phoronix

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

  • videos benchmarks on phoronix

    Hi,

    Phoronix site does a great job in testing graphic cards under *nix, which is very usefull, as those informations doesn't exist out there.

    But the performance part of the tests focus on the comportement of the cards with games. But 3d is not the only use of graphic cards, with democratisation of HTPC (home theater PC) and the HD content waves, video performances are more and more sensible. Especially because we can observe very different comportement between cards due to hardware and drivers properties.

    So I think that adding some video tests (with xvid, avc1, mpeg2, x264 movies) could interest lots of people.

    What do you think about it ?
    21
    Yes, definetly !
    100.00%
    21
    No ( I'am using Windows to play videos )
    0.00%
    0
    Last edited by pouyoux; 03-26-2008, 12:51 PM.

  • #2
    I hope some video benchmarks will be part of the Phoronix Test Suite. Well, they are on my TODO list. Once they are part of the Phoronix Test Suite, there will be video benchmarks appearing in original Phoronix articles.

    The first bits of the Phoronix Test Suite should be out soon and after that anyone is welcome to make any video benchmark scripts, if you beat me to it

    (Pouyoux: Were you the one that emailed me yesterday about video benchmarks? Anyways, my response to that email was going to basically be what's said above. )
    Michael Larabel
    http://www.michaellarabel.com/

    Comment


    • #3
      Great move to "open" the test suite protocol !

      I see 2 parts for the video test :

      1/ performance part : how hard the cpu works to decode some samples in different usual formats at different usual resolutions (576p, 720p, 1080p).
      For this part I could imagine some ways to script it.

      2/ capabilities part : how good does the graphic card do its job.
      Does it provide hardware decoding acceleration by playing some video formats ? is rendering tearing free ? does noise reduction do its job ? does 3:2 pulldown detection do work ? etc etc
      For this part it's a lot more difficult to imagine how to script it

      (Michael : yes I'am tho one who wrote to you yesterday. today I would like to check that I 'm not the only one in *nix world interested by (and using it for) video playback )
      Last edited by pouyoux; 03-25-2008, 09:49 PM.

      Comment


      • #4
        Interesting, I can hardly wait to see some results... by the way, how could performance be assessed for video playback (besides aproximate CPU utilization and XVideo capabilities)?

        Comment


        • #5
          In speaking of brute performance GPU load would be an interesting item, but to my knowledge we can't get it.

          For capabilities, that's not enough that GPU/drivers provide video enhancement, they have to do it well !! And trust me, a good couple GPU/drivers (especially under *nix) is quite rare.

          To illustrate this, you can go there http://www.hqv.com/benchmark.cfm, you have some screenshots, and some pdf which explain deeper what a good couple GPU/drivers have to do to make a good job in video playback.

          HQV test suite exists under serveral formats (PAL DVD, NTSC DVD, BLU-RAY, HD-DVD), maybe DVD version could be used to benchmark video cards ?
          I think BR version need a BR virtual machine to work correctly, so we'll have to wait a few years before using it under *nix

          I don't know how automatic HQV is.
          Last edited by pouyoux; 03-26-2008, 02:31 PM.

          Comment


          • #6
            What a borked poll.

            Comment


            • #7
              FPS with Mplayer

              Hi,
              i have edit a other test-resources to install/start a mplayer decoding benchmark.

              If someone is interested ... but is can be make better.
              Code:
              phoronix-test-suite run mplayer
              
              Would you like to save these test results (Y/n)? n
              
              ====================================
              Mplayer (Run 1 of 3)
              ====================================
              
              FRAMES: 2173.25
              SECONDS: 7.86
              SECONDS INTERLACED: 14.92
              Decoding Rate: 276 fps
              Decoding Rate Deinterlaced: 145 fps
              
              ====================================
              Mplayer (Run 2 of 3)
              ====================================
              
              FRAMES: 2173.25
              SECONDS: 7.53
              SECONDS INTERLACED: 14.03
              Decoding Rate: 288 fps
              Decoding Rate Deinterlaced: 154 fps
              
              ====================================
              Mplayer (Run 3 of 3)
              ====================================
              
              FRAMES: 2173.25
              SECONDS: 7.73
              SECONDS INTERLACED: 14.84
              Decoding Rate: 281 fps
              Decoding Rate Deinterlaced: 146 fps
              
              ####################################
              Mplayer:
              
              
              Average: 0.00 Seconds
              ####################################
              Files http://vdrportal.de/board/thread.php...149#post730149

              - without Option "-vo null" in script video plays in window
              - with "-vc mpeg12" or "-vc xvid" or "-vc ffh264" other codecs would be able

              bye,
              Nando

              Comment


              • #8
                FPS with Mplayer

                Double post can be deleted.
                (I could't read the Message that i have to wait.)
                Last edited by Nando; 06-10-2008, 03:46 PM.

                Comment


                • #9
                  > how could performance be assessed for video playback (besides
                  > aproximate CPU utilization and XVideo capabilities)?

                  Add Xvmc.
                  Measure power consumption.

                  This should be open source software only, nothing binary-only.

                  The quantity numbers should be relatively easy to test.
                  Quality aspects will likely be more difficult.

                  Perhaps multiple player software could also be tested?
                  Mplayer vs xine, vs ...

                  Interlaced context shown on progressing display. Progressive
                  content shown on interlaced display.

                  Some users want/need to build a HTPC to avoid problems
                  found in turnkey solutions:

                  Test the so-called "tricks" modes: freeze frame, slow & fast
                  motion, backwards, etc. The turnkey Ethernet to video bridges
                  are said to do poorly on this.

                  Many turnkey boxes (Tivo, CECBs ...) crash, especially when
                  displaying OTA closed captions.

                  Some HDTVs are said to have poor scalers, so test 480i to 1080i
                  and visa-versa.

                  Comment


                  • #10
                    It would be interesting to compare different video drivers (free & binary) against each other, and then compare different decoders against each other (I'd be interested to see if Fluendo's decoders are worth the money). The decoders could be a problem, as a very large majority of players use libraries from FFmpeg.

                    Encoding might be better suited to a CPU-style test, and if you want to do that, there's not a whole lot better now than the 1080P .pngs from Big Buck Bunny.

                    Comment


                    • #11
                      Originally posted by pouyoux View Post
                      So I think that adding some video tests (with xvid, avc1, mpeg2, x264 movies) could interest lots of people.
                      What do you think about it ?
                      I'd like to second the request for x264 tests. It is a good stress test of an operating systems scheduler because of it's multithreaded nature. It spawns and joins many threads rapidly (currently it doesn't do thread pooling). A recent bug in the linux kernel was discovered and fixed through x264 tests.

                      See http://x264dev.multimedia.cx/?p=185

                      Comment


                      • #12
                        Originally posted by JohnAStebbins View Post
                        I'd like to second the request for x264 tests. It is a good stress test of an operating systems scheduler because of it's multithreaded nature. It spawns and joins many threads rapidly (currently it doesn't do thread pooling). A recent bug in the linux kernel was discovered and fixed through x264 tests.

                        See http://x264dev.multimedia.cx/?p=185
                        A basic x264 test profile can be found in the Phoronix Test Suite Git code for 2.4 Lenvik.
                        Michael Larabel
                        http://www.michaellarabel.com/

                        Comment


                        • #13
                          Hopefully something with really high bitrate - best over 40 mbps with L4.1.

                          Comment


                          • #14
                            Originally posted by Michael View Post
                            A basic x264 test profile can be found in the Phoronix Test Suite Git code for 2.4 Lenvik.
                            Michael, looking at the git diff, I believe your description isn't correct: the VideoLAN project just hosts the git repo & mailing list for x264- it's not their project.

                            Michael, please tell me you did not disable the asm code when compiling x264. Not enabling the ASM in x264 is like ripping the heart out of the program. The assembly is what makes it such a fine program- it's finely tuned to get the absolute most out of processors.
                            The assembly covers x86 (32 bit, 64 bit, and Windows 64 bit); PPC and later ARM versions are also supported in the program.

                            I don't know enough about the options other than to give you the following advice:
                            Examine the presets and see if any of them are suitable- these are what everyone who uses FFmpeg+libx264 to encode will use, and unless they have a very good reason, will use with x264. And ask the authors (especially Dark Shikari) if you want the program to do something in particular- they're quite approachable. They are very happy to work with people to create works that show off what the program can do.
                            Last edited by dashcloud; 11-18-2009, 10:28 PM. Reason: remove some interesting, but not terribly relevant info

                            Comment


                            • #15
                              Originally posted by dashcloud View Post
                              Michael, please tell me you did not disable the asm code when compiling x264. Not enabling the ASM in x264 is like ripping the heart out of the program. The assembly is what makes it such a fine program- it's finely tuned to get the absolute most out of processors.
                              The assembly covers x86 (32 bit, 64 bit, and Windows 64 bit); PPC and later ARM versions are also supported in the program.
                              It was just a temporary limitation until yasm was supported by PTS External Dependencies, which it does now support tonight.
                              Michael Larabel
                              http://www.michaellarabel.com/

                              Comment

                              Working...
                              X