Announcement

Collapse
No announcement yet.

AMD's New Carrizo Graphics PCI IDs

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

  • AMD's New Carrizo Graphics PCI IDs

    Phoronix: AMD's New Carrizo Graphics PCI IDs

    With yesterday's release of the new open-source "AMDGPU" Linux graphics driver stack we finally have a look at some of the hardware enablement code for the graphics processors of the upcoming "Carrizo" APUs...

    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
    Originally posted by Phoronix
    There's also a new H.265 performance hardware decoder exposed for Tonga
    H264.

    Comment


    • #3
      why hardware based encoders

      Originally posted by d2kx View Post
      H264.
      If die-space is really at a premium, why build in a hardware decoder, instead of investing in a software opencl (or vulkan?) based decoder? I understand that such a decoder is atomic and can simple be added on, but couldn't you trade that space out for more cache, or something. An opencl decoder would help everyone, and wouldn't restrict you to a single algorythm - it could easily be upgrade to 26X/vpX as they come out.

      Is it really that big an advantage to be able to process video without any pressure on the CPU/GPU? I guess the 4K dimensions really push the limits.

      Comment


      • #4
        The decoding process involves a mix of activities -- some map well onto parallel engines (shaders, opencl etc..) while others are essentially serial (at least within a slice) and really need specialized hardware to accelerate effectively.
        Test signature

        Comment


        • #5
          Isn't this exact mix of parallel and serial tasks the purpose for HSA? Would it be possible to write a software decoder with good enough performance and a negligible amount of extra power usage compared to the hardware codecs at least on the Kaveri/Carizzo or future APUs?

          Comment


          • #6
            Yes and no -- that fine-grained mix of CPU and GPU is exactly what HSA is intended to improve. Downside is that the serial part of the workload isn't a great fit for CPU either and that's where dedicated hardware comes in.

            The same logic applies to 3D graphics -- a few stages in the graphics pipeline (rasterizer, integrated texture cache/filter etc..) still really benefit from having dedicated hardware and we don't see that going away any time soon.
            Test signature

            Comment

            Working...
            X