system76 / docs

System76 support documentation site
https://support.system76.com
309 stars 400 forks source link

Add info on docker usage to rocm install instructions #1242

Closed jessecambon closed 3 months ago

jessecambon commented 4 months ago

I ran into some permissions issues using Docker Desktop with ROCM and was able to resolve these issues by using docker.io instead and running docker as sudo. You can refer to the comments here for more information.

I thought it would be good to add a note here so that others can avoid these issues.

jacobgkau commented 4 months ago

The terminology for "docker.io" (as opposed to Docker Desktop) is Docker Engine or Docker CE: https://docs.docker.com/engine/install/

I think it'd make sense to provide a link to Docker's installation instructions as an alternative to the Ubuntu-packaged version, since Docker recommends using their own repo: https://docs.docker.com/engine/install/ubuntu/

Aside from that, this looks reasonable. It's expected that Docker commands require sudo; other options such as Podman would allow rootless containers.

jessecambon commented 4 months ago

Makes sense. I haven't tried Docker CE, but I assumed docker io was recommended as it's in the CUDA instructions: https://support.system76.com/articles/cuda

jacobgkau commented 4 months ago

I haven't tried Docker CE,

You are using the Ubuntu package for Docker CE.

jessecambon commented 4 months ago

@jacobgkau I edited the PR per your comments

jessecambon commented 1 month ago

@jacobgkau the current ROCm install instructions are for ROCm version 6.1. I'm curious if you have been able to install version 6.2? In theory we should just be able to edit the file/etc/apt/sources.list.d/rocm.list and point the apt entry to one of the version 6.2 URLs. However, when I do this I run into issues with broken packages (shown below). I'm not sure if this is a ROCm issue or an incompatibility with something in Pop! OS.

Let me know if there's a better place to ask for help on this.

$ sudo apt install rocm --upgrade
Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
Some packages could not be installed. This may mean that you have
requested an impossible situation or if you are using the unstable
distribution that some required packages have not yet been created
or been moved out of Incoming.
The following information may help to resolve the situation:

The following packages have unmet dependencies:
 rocm : Depends: rocm-utils (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: rocm-developer-tools (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: rocm-openmp-sdk (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: rocm-opencl-sdk (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: rocm-ml-sdk (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: mivisionx (= 3.0.0.60200-66~22.04) but 2.5.0.60100-82~22.04 is to be installed
        Depends: migraphx (= 2.10.0.60200-66~22.04) but 2.10.0.60201-112~22.04 is to be installed
        Depends: rpp (= 1.8.0.60200-66~22.04) but 1.8.0.60201-112~22.04 is to be installed
        Depends: rocm-core (= 6.2.0.60200-66~22.04) but 6.2.1.60201-112~22.04 is to be installed
        Depends: migraphx-dev (= 2.10.0.60200-66~22.04) but 2.10.0.60201-112~22.04 is to be installed
        Depends: mivisionx-dev (= 3.0.0.60200-66~22.04) but it is not going to be installed
E: Unable to correct problems, you have held broken packages.
jacobgkau commented 1 month ago

@jessecambon Why did you specify --upgrade in your apt install command?

I'm getting a different error from you when trying to install version 6.2.0 with 6.2 in the URL:

The following packages have unmet dependencies:
 rocdecode: Depends: mesa-amdgpu-va-drivers but it is not installable

I get the same message with 6.2.2 or latest in the URL. Did you deviate from the System76 instructions and manually install a newer amdgpu-install .deb package from the AMD instructions?

jessecambon commented 1 month ago

@jacobgkau --upgrade probably wasn't necessary in that command. I didn't install amdgpu-install, but I did get the same mesa-amdgpu-va-drivers error message you are getting when running sudo apt install rocm with 6.2.2 and latest. The error message I showed above was from using 6.2.