No announcement yet.

AMD Radeon ProRender Plug-Ins Switch To HIP For Newer GPUs Rather Than OpenCL

  • Filter
  • Time
  • Show
Clear All
new posts

  • AMD Radeon ProRender Plug-Ins Switch To HIP For Newer GPUs Rather Than OpenCL

    Phoronix: AMD Radeon ProRender Plug-Ins Switch To HIP For Newer GPUs Rather Than OpenCL

    Back in March the AMD Radeon ProRender SDK 3.1 finished transitioning to HIP for using AMD's HIP C++ interface rather than OpenCL for newer Radeon GPUs. Following that, the Radeon ProRender plug-ins have also now finished being converted over to using HIP on AMD Radeon Vega (GFX9) GPUs and newer...

    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
    From a business perspective this makes sense. Not so many consumers running that software.

    TL;DR Sad to see AMD moving away from OpenCL when they made such a big push for it.

    People that just want a stable API that works on all their devices. I must have been ignorant thinking OpenCL was just that...

    Turns out (ironically) that it's not AMD's fault for that. AMD were the ones doing their best to make OpenCL work and Nvidia / Khorons (intentionally or not) decided to stab OpenCL (and everyone using it) in the back. It's sad to see AMD moving away from OpenCL. I and many others have made it abundantly clear that we don't like ROCM/HIP.

    I've started using Vulcan-compute and DirectML via WSL2 for my compute needs. I didn't expect things to be easier going the following route to get... Linux / KVM -> Windows11* / WSL2 -> ... finally a stable and reliable Linux compute environment!!!!!! (* Windows 10 doesn't support nested VMs in WSL2 for AMD systems)

    It's absolutely mad. I honestly cannot understand how anybody would waste their time doing that. I at least know KVM like the back of my hand and had the system configured and optimized for gaming. If anyone wanted to know why every single Windows gaming CS student was forced to move to Nvidia as trash as their cards are, there's your reason.

    I _REALLY_ don't like using DirectML but hey it actually works and not only that but it works on the following GPUs:

    AMD GCN 1st Gen (Radeon HD 7000 series) and above
    Intel Haswell (4th-gen core) HD Integrated Graphics and above
    NVIDIA Kepler (GTX 600 series) and above
    Qualcomm Adreno 600 and above

    ‚ÄčI'm still rooting for rusticl. It's clear that it's up to an open source project to solve basic Linux compute. Just not sure if Linux will get good compute before Java makes a comeback.


    • #3
      Jabberwocky I'm not inclined to agree. AMD had many opportunities to really help OpenCL and did the bare minimum. CUDA, as far as I understand, wasn't inherently better for most of its existence. It was popular because Nvidia really went out of their way to make it easy to adopt. They poured a lot of money into building libraries, making good documentation, providing useful demos, and working directly with 3rd party developers. I don't recall AMD doing any of that. As much as I avoid Nvidia where possible, CUDA really did earn its popularity.

      There was a hobby project I started a few years ago depending on OpenCV. I really wanted to use OpenCL but I simply could not be bothered with the lack of documentation, incomplete drivers, and over-complicated setup. I spent hours just trying to figure out how to make it do anything at all and understand what it is I'm doing, yet within a half hour of trying CUDA for the first time, I already got usable results and I had a decent grasp of what each line of code did.

      Should OpenCL ever be successful, it'll be thanks to Intel, who actually seems to understand what it takes to outpace CUDA. Unfortunately for Intel, they don't have the marketshare to influence change, at least not yet. So it may take a few years until there is compelling competition against CUDA.