Thanks for the writeup.
However, this is still using the (closed source) AMDGPU-Pro driver on top of the (open source) AMDGPU Kernel driver. In this thread we already stated that that’s possible, but also discussed why that might not be a good option.
It will help people that can use the Pro driver, but it seems there is no solution for people that can’t or don’t want to run the Pro driver.
do the instructions linked also work for nvidia drivers?
AFAIK for nvidia it should “just work”. The driver is completely different, but if the ICDs are the issue then probably yes.
Hey @mihawk90 ! I’m getting the same issue, my Vega 64 is detected but OpenCL Not detected: clGetDeviceIDs(). I don’t care about gaming performance because I don’t game.
Any progress on your end so far?
For AMD you need the opencl bits from the amdgpu-pro driver for functioning opencl, or install ROCm if the project you want to run supports it and not everything does. But ROCm isn’t supported on Navi for those that have one. ROCm and amdgpu-pro are two different things and neither one requires the other.
For nvidia you just need to install cuda-toolkit for functioning opencl. Opencl on Nvidia is easy, on AMD it’s a train wreck. I’m currently running both.
In that case just use the AMDGPU-Pro driver from AMD.
@mihawk90 from here https://www.amd.com/en/support/kb/release-notes/rn-radpro-lin-16-40?
What are the proper steps to take so I don’t break things at this point? >_<
Yes that one. I never installed/used it but if you’re also using Fedora you can use the driver for CentOS/RHEL and it should "just work"™.
Although if you’re unlucky the Pro driver doesn’t work with the newer Kernels in Fedora since both CentOS and RHEL are on LTS Kernels.