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
[2] https://wiki.archlinux.org/title/AMDGPU
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.
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
[2] https://github.com/intel/compute-runtime/issues/469#issuecom...
[3] https://rocmdocs.amd.com/en/latest/Installation_Guide/Instal...
this may become the case once hardware is good for 20 years. we aren't there yet, though we're closer than a decade ago. (typing this on an aging 8 year old desktop, which is really due for an upgrade).