Announcement

Collapse
No announcement yet.

POCL Is Working On OpenCL-Over-CUDA With New Backend

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

  • POCL Is Working On OpenCL-Over-CUDA With New Backend

    Phoronix: POCL Is Working On OpenCL-Over-CUDA With New Backend

    POCL, the Portable Computing Language project, aiming to provide a portable OpenCL implementation for usage against different CPUs and back-ends, has been working on CUDA support...

    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
    That's really good news! Such an early state and they are already exceeding the performance of nvidias opencl driver in some benchmarks.

    Comment


    • #3
      why isnt the nvidia driver already doing this?

      Comment


      • #4
        Originally posted by cj.wijtmans View Post
        why isnt the nvidia driver already doing this?
        Good OpenCL support would enable competition. Now everyone uses CUDA which is exclusively for NVIDIA.

        Comment


        • #5
          Originally posted by caligula View Post

          Good OpenCL support would enable competition. Now everyone uses CUDA which is exclusively for NVIDIA.
          True, but as a result, we only have CUDA code. A CUDA over OpenCL translation layer would be needed as well :/

          A good thing with vulkan is that the SPIR-V shades are shared with OpenCL.

          Comment


          • #6
            Originally posted by M@yeulC View Post
            True, but as a result, we only have CUDA code. A CUDA over OpenCL translation layer would be needed as well :/
            It's not like you are going to run CUDA software on Intel's GPU so, here is solution for running CUDA code on AMD: https://github.com/GPUOpen-Professio...pute-Tools/HIP

            Comment


            • #7
              Beyond allowing people to run applications using OpenCL versions that aren't supported on particular hardware, I don't understand what the point of this is. Isn't OpenCL in itself supposed to be highly portable API allowing applications to run on a myriad of CPUs, GPUs, DSPs and FPGAs?

              Comment


              • #8
                Originally posted by M@yeulC View Post

                True, but as a result, we only have CUDA code. A CUDA over OpenCL translation layer would be needed as well :/

                A good thing with vulkan is that the SPIR-V shades are shared with OpenCL.
                AFAIK there was a CUDA --> HSA translator thing, "boltzmann initiative" was it?

                Comment

                Working...
                X