Announcement

Collapse
No announcement yet.

OpenCL 3.0 Bringing Greater Flexibility, Async DMA Extensions

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

  • OpenCL 3.0 Bringing Greater Flexibility, Async DMA Extensions

    Phoronix: OpenCL 3.0 Bringing Greater Flexibility, Async DMA Extensions

    OpenCL 3.0 is being released today in provisional form. OpenCL 3.0 is making OpenCL 2.x functionality now optional to make it better suited for a range of devices plus there is new functionality introduced like subgroups, extensions for asynchronous data copies, and more.

    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
    Nice, NVIDIA can dictate whatever they want to Khronos and Khronos has to comply like a good lapdog. Seriously, what has been so hard to implement from OpenCL 2.x all these years? This is just an excuse by Nvidia in order to push for CUDA and damage OpenCL. Don't tell me they don't provide similar functionality on CUDA.... I am not a gpgpu developer and i haven't looked much into it, but i highly doubt CUDA doesn't offer something similar to SVM. Hell, even Nouveau could implement this.

    This is seriously sick. A whole "independent" "open standards" organization getting pushed around by a single company. And the same applies to Microsoft, how many times has Nvidia pushed Microsoft around regarding D3D? Hell even D3D12 was split into two versions (12_0 and 12_1), with initially AMD supporting the real d3d12 (12_0) and Nvidia NOT SUPPORTING it fully in hardware but supporting some other minor additions (and calling them 12_1). And you had clowns for years claiming Nvidia has better D3D support because "they support D3D12.1, much better than AMD".

    OpenCL 3.0 is the same. They literally threw OpenCL 2.x in the garbage bin (optional features mean almost no one will use them) just because Nvidia refused to support it in order to push CUDA. Good job Khronos.

    Comment


    • #3
      It's kind of sad and funny to think that the lack of progress means that Apple could basically change their OpenCL 1.2 driver to report as OpenCL 3.0 and they'd actually be up to date with a Khronos standard after all this time.

      Comment


      • #4
        Really? o-o

        Apple, your comments on this matter?

        Comment


        • #5
          to be honest i never see any real use for opencl at all

          Comment


          • #6
            I mean, OK good to see OpenCL 3.0 releases. But who supports it? AMD may get OpenCL 3.0 on ROCm is a few months, following Intel. But NVIDIA won't nor all mobile vendors.

            Comment


            • #7
              Originally posted by TemplarGR View Post
              Nice, NVIDIA can dictate whatever they want to Khronos and Khronos has to comply like a good lapdog. Seriously, what has been so hard to implement from OpenCL 2.x all these years? This is just an excuse by Nvidia in order to push for CUDA and damage OpenCL. Don't tell me they don't provide similar functionality on CUDA.... I am not a gpgpu developer and i haven't looked much into it, but i highly doubt CUDA doesn't offer something similar to SVM. Hell, even Nouveau could implement this.
              David Airlie makes some comments about how the situation is a bit more complicated than you suggest. Watch https://www.youtube.com/watch?v=KfDQb6xOkXg, around the 12~13 minute mark...

              Comment


              • #8
                I find it quite funny that an OpenCL 2.0 driver could support more features than an OpenCL 3.0 driver.

                That said, I like the new, more modular approach to this. Though I have 0 experience with GPGPU or OpenCL, so my opinion on this is quite likely irrelevant ¯\_(ツ)_/¯.

                Comment


                • #9
                  Originally posted by marty1885 View Post
                  I mean, OK good to see OpenCL 3.0 releases. But who supports it? AMD may get OpenCL 3.0 on ROCm is a few months, following Intel. But NVIDIA won't nor all mobile vendors.
                  Actually, since ver 3.0 basically makes everything optional that Nvidia doesn't support, it would be quite easy for them to support OpenCL 3.0.

                  Comment


                  • #10
                    Originally posted by lucasbekker View Post

                    David Airlie makes some comments about how the situation is a bit more complicated than you suggest. Watch https://www.youtube.com/watch?v=KfDQb6xOkXg, around the 12~13 minute mark...
                    Thanks for the link, it seems a Mesa Level 0 is his new idea for going forward.

                    Comment

                    Working...
                    X