Announcement

Collapse
No announcement yet.

Faster VP9 Decoding Is On The Horizon

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

  • Faster VP9 Decoding Is On The Horizon

    Phoronix: Faster VP9 Decoding Is On The Horizon

    For those frustrated by the current lack of hardware supporting VP9 encode/decode and the slow decode speed when playing back VP9 content on the CPU, improvements are coming...

    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
    Only 30% faster? Then it's still far behind ffmpeg: https://raw.githubusercontent.com/rb.../vp9-32bit.png

    Comment


    • #3
      *cough* OpenCL GPU decoder *cough*
      (Just saying.)

      Comment


      • #4
        Originally posted by plonoma View Post
        *cough* OpenCL GPU decoder *cough*
        (Just saying.)
        Can the myth about de- and encoding on the GPU ALUs now die? Thats not effective. You cant do everything on the GPUs.

        Comment


        • #5
          Originally posted by Nille View Post
          Can the myth about de- and encoding on the GPU ALUs now die? Thats not effective. You cant do everything on the GPUs.
          Indeed. In particular the CABAC (Context-adaptive binary arithmetic coding) process is an inherently serial process, and is extremely slow to do on a GPU which requires taking advantage of parallelism to be fast. CABAC takes a large portion of the time required for h264 and 265.

          Comment


          • #6
            Originally posted by Nille View Post
            Can the myth about de- and encoding on the GPU ALUs now die? Thats not effective. You cant do everything on the GPUs.
            Yes, the point of OpenCL is that you can do anything on the GPUs.

            Comment


            • #7
              Originally posted by eydee View Post
              Yes, the point of OpenCL is that you can do anything on the GPUs.
              Now, of course it's more efficient to have dedicated hardware acceleration, but I find it more efficient to be able to drop the encoder/decoder anywhere and have it work on a standard that 90% of everything supports.

              Comment


              • #8
                Originally posted by eydee View Post
                Yes, the point of OpenCL is that you can do anything on the GPUs.
                Why do you thing that. Did you ever ask yourself, why software is not scaling with the number of CPU cores? You cant do most of the work parallel and thats a hard requirement on GPUs.

                Comment


                • #9
                  Originally posted by eydee View Post
                  Yes, the point of OpenCL is that you can do anything on the GPUs.
                  In theory you can maybe, but there's no point doing in practice when it'll be less efficient than doing it on the CPU. GPUs are good at *massive* parallelism. Video decoding is not massively parallel. The entropy coder is extremely serial, and even other parts are so interdependent that they do not lend themselves to massive parallelism.

                  Comment


                  • #10
                    this means SSE instructions are not involved?! ahahahah What part of CPU process VP9 decode? FPU!?

                    Comment

                    Working...
                    X