But the built in Linux drivers usually don’t work well on Einstein from what I’ve seen. Try installing the ROCM drivers or the driver installer from AMD’s website for your hardware.
Unfortunately Fedora is not supported by AMD ROCm. But I tried CentOS Stream 9 with ROCm package on the same machine, and then the ASGWS tasks are executed correctly.
the reality is that you might not be able to. AMD drivers are a bit of a mess on linux with inconsistent support across distros and kernel versions.
if you only want to use FC41 and don't have any luck installing from the amdgpu-install script from the AMD drivers download site, you might want to go over to the fedora dev spaces (i dont know where these are) or support forums to find out where you need to submit bug reports to fix it.
It definitely makes sense to look for help from development teams. But first I need to collect enough pertinent information to be able to articulate the issue properly and address the right team.
The best way to do this is to reproduce the issue locally.
I wonder whether some kind of troubleshooting guide exists to track down such failure root cause.
If all you want to do with your GPU is run Einstein, then I would just install rocminfo, rocm-opencl, and rocm-clinfo. I would also recommend installing rocm-smi, which is a monitoring and control app for AMD GPUs.
*Edit Your NAVI GPU may or may not work depending on how Fedora builds and packages ROCm, since NAVI is not officially supported.
Actually the AMD GPU was working, more or less, fine until I upgraded the FC40 to FC41;
All the required packages were installed correctly. But I struggled against a kernel issue for a long time.
Finally I gave up with AMD GPU and replace it with NVIDIA GPU.
The NVIDIA GPU based opencl package is much more reliable and mature. However there are some occasional faults happen that I have not seen with AMD GPU before. This is a topic for another discussion.
But the built in Linux
)
Unfortunately Fedora is not supported by AMD ROCm. But I tried CentOS Stream 9 with ROCm package on the same machine, and then the ASGWS tasks are executed correctly.
My aim is to get FC41 running correctly however.
the reality is that you might
)
the reality is that you might not be able to. AMD drivers are a bit of a mess on linux with inconsistent support across distros and kernel versions.
if you only want to use FC41 and don't have any luck installing from the amdgpu-install script from the AMD drivers download site, you might want to go over to the fedora dev spaces (i dont know where these are) or support forums to find out where you need to submit bug reports to fix it.
_________________________________________________________________________
@Ian&Steve C. Thanks for
)
@Ian&Steve C.
Thanks for your replies!
It definitely makes sense to look for help from development teams. But first I need to collect enough pertinent information to be able to articulate the issue properly and address the right team.
The best way to do this is to reproduce the issue locally.
I wonder whether some kind of troubleshooting guide exists to track down such failure root cause.
Fedora packages ROCm and it
)
Fedora packages ROCm and it is packaged for FC41, so all you should have to do is install the relevant packages. https://fedoraproject.org/wiki/SIGs/HC#Installation
If all you want to do with your GPU is run Einstein, then I would just install rocminfo, rocm-opencl, and rocm-clinfo. I would also recommend installing rocm-smi, which is a monitoring and control app for AMD GPUs.
*Edit Your NAVI GPU may or may not work depending on how Fedora builds and packages ROCm, since NAVI is not officially supported.
@tictoc: Thanks for you
)
@tictoc: Thanks for you response!
I have noticed it just now.
Actually the AMD GPU was working, more or less, fine until I upgraded the FC40 to FC41;
All the required packages were installed correctly. But I struggled against a kernel issue for a long time.
Finally I gave up with AMD GPU and replace it with NVIDIA GPU.
The NVIDIA GPU based opencl package is much more reliable and mature. However there are some occasional faults happen that I have not seen with AMD GPU before. This is a topic for another discussion.