Announcement

Collapse
No announcement yet.

Blender 3.1 Released With New Features Sans AMD HIP Linux GPU Acceleration

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

  • #21
    Originally posted by darkbasic View Post
    That's not the API fault. It was garbage compared to CUDA because all development went towards the CUDA API.
    I should have added the word "practically", yes.

    Comment


    • #22
      Originally posted by darkbasic View Post

      That happens when all your development is towards a proprietary API and OpenCL gets the second tier citizen treatment.
      not even true and you can verify it for yourself. go clone the source, then look at the git logs. OpenCL got way more effort than it is worth.

      OpenCL is an API both unsuitable for blender. and had trash support on be much the only cards it was used on. only really intel, and AMD used it. intel was a little better than amd, but it still wasn't good. and opencl drivers are quite frankly poor across the board.

      openCL was a poor fit from blender right from the beginning and was implemented out of necessity. that necessity is gone, they're working on Vulkan and AMD support.

      OCL got first class citizen work done on it. and only ever produced third class results. that's why it was removed.

      Comment


      • #23
        Originally posted by darkbasic View Post

        That's not the API fault. It was garbage compared to CUDA because all development went towards the CUDA API.
        Absolutely Wrong. Borderline Fallacy.

        It was a combination of a poorly thought out API (openCL 1.0) and an incomplete driver implementation (Register spilling was never properly implemented in the SDK nor the driver, and that's the main reason big kernels fails, like the original monolithic Cycles kernel engine).

        Cycles (Blender's raytracer engine) was developed with OpenCL, CUDA and Software rendering in it's first version. CUDA and software worked, but not OpenCL. So OpenCL was left as a WIP, until the OpenCL compiller were able to succesfully compille the code, and that's happened when AMD implemented OpenCL 1.2, but then again, a mixture of hardware limitations and driver bugs prevented it to be a viable alternative on AMD, except on their more expensive cards at the time (Only Graphics Core Next Cards were confirmed to work with Cycles OpenCL megakernel implementation for simple scenes, the previous terascale architecture proved to be insufficient for OpenCL 1.2, altough AMD tried it best to support OpenCL 1.2 on that architecture.). For reference, the OpenCL Cycles megakernel worked without problems or crashes on nvidia cards, but it was slower than it's CUDA counterpart.

        Comment

        Working...
        X