Drivers & Software

[Meteorhead]

Because the ROCm stack even with the coming 1.8-beta does not support Ubuntu 18.04 I went ahead and tried AMDGPU(-PRO) 18.20 as a last resort before segmenting the homogenous cluster and downgrading AMD nodes to 17.10 or even remaining on 16.04. After installing the driver packages manually (because the installer script fails due to missing dependencies of i386 packages (why is there not option to omit 32-bit support on a 64-bit node??)), the module amdgpu fails to load. kern.log is attached.

How should the drivers be installed, because the "Linux installation guide" does not work.

[riuku]

Try adding the following commands prior to installing amdgpu drivers

sudo su

dpkg --add-architecture i386; apt-get update

this should allow you to finish the instillation

[Meteorhead]

Sorry for the late reply. Indeed, it did solve the problem.

[kingfish]

Why don't you mark the reply as "solved"/correct answer so others may be helped?

[Meteorhead]

Because I don't see the mark to have it answered. I must've opened a discussion, not a question.