Tim Gu

@jinz2014 sure! Just a note that we would prefer closing issues with no further actionable items to help better track which ones still require active attention. Since the feature requested...

Closing the issue since the commit in now part of ROCm 6.3.x.

@AngryLoki, Thanks for reaching out! Unfortunately, we do not currently have plans to bump to C++ 17 standard, and likely will not unless the decision is made for all math...

Hi, I will be closing this issue following the merge of #568. Thanks!

Hi @lovely-error thank you for reaching out! By sometimes, do you mean that the error is transient? Also can you double check to see if runtime power management is enabled...

@lovely-error seems like `PM: not prepared for power transition: code -12` suggest that the system is out of memory. Are you running any memory-heavy jobs by any chance? If not...

@lovely-error Thank you for the additional context. To help us better diagnose your problem would you mind running ``` apt show amdgpu ``` And paste the exact outputs? Thanks!

> For anyone in need of an easy way to reproduce this problem: > > I'm using Ollama, and I consistently have this issue whenever a model is running on...

@djpetti This might be related to your issue (https://gitlab.freedesktop.org/drm/amd/-/issues/2362). It should've been fixed in the latest driver with this commit. https://github.com/ROCm/ROCK-Kernel-Driver/commit/5095d5418193eb2748c7d8553c7150b8f1c44696. Can you please verify that there is enough ram/swap...

@lovely-error @djpetti It will also help if you can set `/sys/power/pm_debug_messages` to 1 and show the logs from dmesg again. It should reveal more details.