5 ms·
The FOSS community depends on contributions from GPU vendors to have their products supported on Linux in a reasonable timeframe. As long as the contributions a
by commoner 5y ago
The FOSS community depends on contributions from GPU vendors to have their products supported on Linux in a reasonable timeframe. As long as the contributions are released under a FOSS license, the vendors are part of the FOSS community.
Intel provides excellent Linux support for its GPUs through open source drivers,[1] and is about to launch a line of dedicated GPUs in early 2022 with the same level of support.
AMD has provided good support for essential GPU features through its open source Linux driver since 2015,[2] but compute features such as full OpenCL support (for most AMD models) are still locked in its proprietary drivers.[3]
Nvidia's open source Linux driver contributions are minimal, and they've earned a bad reputation for that. The reverse-engineered open source Nouveau driver is an incredible effort, but falls behind Nvidia's proprietary driver in performance and feature support.[4] This is what happens when the hardware vendor doesn't cooperate with the FOSS community.
[1] https://wiki.archlinux.org/title/Intel_graphics https://wiki.archlinux.org/title/Intel_graphics
[2] https://wiki.archlinux.org/title/AMDGPU https://wiki.archlinux.org/title/AMDGPU
[3] https://wiki.archlinux.org/title/AMDGPU_PRO https://wiki.archlinux.org/title/AMDGPU_PRO
[4] https://wiki.archlinux.org/title/Nouveau https://wiki.archlinux.org/title/Nouveau
- schmorptron 5y agoSo does Intel also include their compute features in the mainline drivers, unlike AMD? That'd be something to push me towards them when GPUs become available and somewhat reasonably priced again
- commoner 5y agoIntel's compute features are in a separate open source package called Intel Graphics Compute Runtime.[1] Integrated graphics from 2014 (Broadwell) and later are supported. Support for discrete graphics (2020 onward) is in development.[2] AMD provides an open source OpenCL solution for Linux called ROCm, but the project is too limited to be helpful in many use cases. ROCm supports a total of 6 GPU models, all from previous generations,[3] and does not support GUI-based software applications such as Blender.[4] [1] https://github.com/intel/compute-runtime https://github.com/intel/compute-runtime [2] https://github.com/intel/compute-runtime/issues/469#issuecomment-964924940 https://github.com/intel/compute-runtime/issues/469#issuecom... [3] https://rocmdocs.amd.com/en/latest/Installation_Guide/Installation_new.html#confirm-you-have-a-rocm-capable-gpu https://rocmdocs.amd.com/en/latest/Installation_Guide/Instal... [4] https://github.com/RadeonOpenCompute/ROCm/issues/1397 https://github.com/RadeonOpenCompute/ROCm/issues/1397
- schmorptron 5y agoAwesome, thanks!
- mijoharas 5y agoI've got to agree, nvidia have lost any love I might have had for them, and I'm definitely not going to be buying another product from them any time soon. I'm still very impressed with the great work the nouveau people are doing to work around nvidia's stubbornness though, so congratulations to them.