https://bugs.freedesktop.org/show_bug.cgi?id=107141
--- Comment #4 from Mihai Preda mhpreda@gmail.com --- Please see my original report and discussion on ROCm: https://github.com/RadeonOpenCompute/ROCm/issues/605
Basically, the manually configured sclk is reset (lost) on monitor state change ("monitor on").
As per @kentrussell's feedback there, I'm reporting this as a suspected amdgpu issues related to power management.
I can reproduce with a recent 5.0 kernel and most recent ROCm (2.2).
~$ uname -a Linux x2 5.0.2-050002-generic #201903131832 SMP Wed Mar 13 22:35:19 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux
apt list --installed | grep rocm
WARNING: apt does not have a stable CLI interface. Use with caution in scripts.
rocm-opencl/Ubuntu 16.04,now 1.2.0-2019030702 amd64 [installed] ~$ apt list --installed | grep hsa
WARNING: apt does not have a stable CLI interface. Use with caution in scripts.
hsa-ext-rocr-dev/Ubuntu 16.04,now 1.1.9-55-gbac2a9b amd64 [installed,automatic] hsa-rocr-dev/Ubuntu 16.04,now 1.1.9-55-gbac2a9b amd64 [installed,automatic] hsakmt-roct-dev/Ubuntu 16.04,now 1.0.9-121-g876627e amd64 [installed,automatic] hsakmt-roct/Ubuntu 16.04,now 1.0.9-121-g876627e amd64 [installed,automatic]
I have reproduced this bug on a wide range of AMD GPUs. Most recently on AMD Vega64.