Announcement

Collapse
No announcement yet.

Dav1d 0.5 Released With AVX2, SSSE3 & ARM64 Performance Improvements - Benchmarks

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

  • Dav1d 0.5 Released With AVX2, SSSE3 & ARM64 Performance Improvements - Benchmarks

    Phoronix: Dav1d 0.5 Released With AVX2, SSSE3 & ARM64 Performance Improvements - Benchmarks

    Friday marked the release of dav1d 0.5 as the newest version of this speedy open-source AV1 video decoder. With dav1d 0.5 are optimizations to help out SSSE3 most prominently but also AVX2 and ARM64 processors. Here are some initial benchmarks so far of this new dav1d video decoder on Linux...

    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
    Looks like there are some problems with performance on Threadripper with 4K material. Maybe infinity fabric is maxed out and can't work efficiently or something.

    Comment


    • #3
      After this, do you think Google should continue with libgav1...?

      Comment


      • #4
        Originally posted by tildearrow View Post
        After this, do you think Google should continue with libgav1...?
        AV1 is a new video codec that aims to provide improved compression efficiency compared to VP9 and H.265/HEVC. Over the past year, we’ve…


        The Performance is not yet good with libgav1 compared to dav1d but i think google will do what they want

        Comment


        • #5
          4k 10 bit is notably missing

          Comment


          • #6
            Originally posted by pal666 View Post
            4k 10 bit is notably missing
            Not sure if there are already 4k 10bit AV1 Videos to test with and as long as the decoder like dav1d get with strong machines 1080p 10bit only a little bit more than 60fps

            "They still seek funding for 10-bit development. And actually, the dav1d team expects 10-bit and 12-bit performance to become almost identical, many assembly functions will even use the same code."

            https://www.reddit.com/r/AV1/comment...m_medium=web2x
            Last edited by Toggleton; 12 October 2019, 02:26 PM.

            Comment


            • #7
              Originally posted by tildearrow View Post
              After this, do you think Google should continue with libgav1...?
              I believe they should.
              Intel launched this project to be optimised for 'amd64'( and maybe thinking only in Intel own processors.. ).

              Google understands that it needs also a project like this for 'aarch64', its not only phones, chrome books are out there, running on arm, GNU/Linux on arm ..
              • dav1d is optimised for 'amd64'
              • libgav1 is optimised for aarch64( its understandable that google choose to go with SSE4.1 for 'amd64', since its the basic requirement for the arch, and also not the default target..default is 'aarch64' ).
              So, in my opinion, they are not direct competitors..

              EDIT: "Intel launched this project"... not this project but SVT-AV1
              Last edited by tuxd3v; 12 October 2019, 08:05 PM. Reason: errata

              Comment


              • #8
                Originally posted by tuxd3v View Post
                [*]dav1d is optimised for 'amd64'
                That's a lie. dav1d is optimized for amd64, i386, armv7 and aarch64 with initial support for powerpc64le too.

                Originally posted by tuxd3v View Post
                [*]libgav1 is optimised for aarch64( its understandable that google choose to go with SSE4.1 for 'amd64', since its the basic requirement for the arch, and also not the default target..default is 'aarch64' ).
                It's nothing but complete overlap. It's stupid, but not surprising from Google.

                Originally posted by tuxd3v View Post
                So, in my opinion, they are not direct competitors..
                You're clueless.

                Comment


                • #9
                  Originally posted by tuxd3v View Post
                  I believe they should.
                  Intel launched this project to be optimised for 'amd64'( and maybe thinking only in Intel own processors.. )
                  So, in my opinion, they are not direct competitors..
                  maybe you're confusing it with https://github.com/OpenVisualCloud/SVT-AV1 that has a decoder but the main focus is the encoder part and is x86_64 only and by Intel

                  and if you look at tests of libgav1 vs Dav1d on aarch64 like this https://docs.google.com/spreadsheets...gid=1359401180 you see that dav1d is a lot faster (Yet). Same results on my Odroid N2 https://code.videolan.org/videolan/d.../15#note_44873 (libgav1 results are 2 posts later... the results are FPS)
                  Last edited by Toggleton; 12 October 2019, 07:26 PM.

                  Comment


                  • #10
                    Originally posted by brad0 View Post
                    That's a lie. dav1d is optimized for amd64, i386, armv7 and aarch64 with initial support for powerpc64le too.
                    It's nothing but complete overlap. It's stupid, but not surprising from Google.
                    I think you need to calm down..
                    Remember that google has tones of devices running Android til version 9 without support.. so libgav1, was maybe a answer to that..

                    maybe dav1d was "too much" focused in 'amd64', maybe with that Google realised that it needed to launch its project..
                    I don't know, and probably only the people that decided to launch the project knows for sure..

                    Anyway, whats the problem for Google launching libgav1?? Your reaction seems strange in a pure OpenSource context..

                    In some time will appear hardware for av1, and this SIMD codecs( burning lots of energy ) will turn themselves irrelevant for new devices..

                    Comment

                    Working...
                    X