Did you consider trying ROCm?
https://github.com/RadeonOpenCompute/ROCm
(in my oppinion it's at least on par with amdgpu-pro performance-wise)
It may be interesting to see if it encounters the problem in the same way.
(OTOH this may be too much trouble just to debug this issue).
Quote:
Originally Posted by SELROC
I am doing another test and waiting for more time before shutting down the system, to see if any other messages are generated in dmesg.
|