Announcement

Collapse
No announcement yet.

Dav1d 0.7 Speedups Are Looking Great On Various Intel + AMD CPUs

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

  • Dav1d 0.7 Speedups Are Looking Great On Various Intel + AMD CPUs

    Phoronix: Dav1d 0.7 Speedups Are Looking Great On Various Intel + AMD CPUs

    This week marked the release of the dav1d 0.7 AV1 video decoder with more performance optimizations thanks to more hand-tuned Assembly and other tweaking of this leading CPU-based AV1 video decoder. Here are benchmarks compared to the prior dav1d 0.5 and 0.6 releases...

    http://www.phoronix.com/scan.php?pag....7-Performance

  • #2
    Sadly, the 10-bit decoding speed has not increased. I have several videos stored in 10-bit AV1, hope I can watch them without dropping frames some time.

    Comment


    • #3
      I've tossed AV1 for now. I am ripping my Blu-ray library with libvpx (VP9).

      Comment


      • #4
        the minimums are exciting :+ )

        Comment


        • #5
          It will be kind if dev1d developer don't use pure assembler and they start to use intrinsics

          Comment


          • #6
            I liked the comment on the dev blog :
            In the future, getting faster on those Intel CPU is going to be very difficult (I know I said that already many times, but this time it's true).

            And i also use VP9 for encoding (well sometime i prefer keeping the video/audio of my blu-ray untouched), because i'm not really fan of the patent part of the H.264/HEVC, but it's really slow compared to H.264.

            Comment


            • #7
              Now please benchmark encoding in rav1e, aom and Intel svt. Looks like rav1e is now in very good stare, even OpenMandriva adapt it as default encoder.

              Comment


              • #8
                Originally posted by Eraserstp View Post
                Sadly, the 10-bit decoding speed has not increased. I have several videos stored in 10-bit AV1, hope I can watch them without dropping frames some time.
                It's only an early version, I don't expect they are optimizing everything at the same time.
                Originally posted by Ski-lleR View Post
                I liked the comment on the dev blog :
                In the future, getting faster on those Intel CPU is going to be very difficult (I know I said that already many times, but this time it's true).
                I'm not sure exactly which CPUs he was talking about, but it looks like there's remaining work to be done on Ice Lake. Just so when, you know, we can actually buy them, performance will be in line with everything else

                Comment


                • #9
                  Very interesting. It might be interesting to see the same benchmarks run on that new AMD laptop you have. I don't expect stunning victories out of the laptop but it is obvious that AMD has come a long way with its laptop chips. So curiosity.

                  Edit:
                  I just checked my Fedora release, they are still on 5.2
                  Last edited by wizard69; 05-24-2020, 04:07 AM.

                  Comment


                  • #10
                    Originally posted by miskol View Post
                    It will be kind if dev1d developer don't use pure assembler and they start to use intrinsics
                    I think it was said that Hand written assembly is faster than intrinsics. So i don't think it makes sense to use intrinsics for dav1d.
                    And a lot assembly is done already
                    With 0.7.0 release, the assembly for x86 CPUs (32bit and 64bit) is now totally complete for the 8bit bitdepth. Source http://www.jbkempf.com/blog/post/202...0-mobile-focus
                    10bit+12bit has no assembly on x86 yet .

                    And 8+10+12bit is in a nice state on arm64 too.

                    Benchmark of a ryzen 3600 and a odroid N2(arm64 Quad-core Cortex-A73(1.8Ghz) and Dual-core Cortex-A53(1.9Ghz))
                    https://openbenchmarking.org/result/...NI62&obr_sor=y
                    and the Arm cpu(using 11Watt) is on 10bit as fast as i7-8565U and i7-1065G7

                    Comment

                    Working...
                    X