Skip to main content ↑ ↓
Topic solved
This topic has been marked as solved and requires no further attention.
Topic: [SOLVED] Strange Observation: kernel 5.11.* + AMDGPU (Read 786 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

[SOLVED] Strange Observation: kernel 5.11.* + AMDGPU

Good Evening,

Artix continues to impress and run very, very well. Just like Linux should....OpenRC is simply great.....

My post below is NOT an Artix issue - it is a kernel issue. And the issue is not show-stopping....I don't really want anyone to burn any time on this - I'm simply looking for someone more in the know than I am regarding kernel internals as to why.....

No bugs along these lines that I see filed - so I filed one @ kernel.org.

AMD Threadripper Gen-I 16 c, 32T, 64gb DDR4 ram, RX-480 (Ellesmere) GPU.

When I run any 5.11.* kernel, my GPU temp is 40 degF hotter than any previous kernel. This is true if I'm running a 5.11.*-Artix, or a 5.11.*-Zen kernel.

Observed temperatures for C++ development and general office type workloads:

5.10.*(Artix or Zen) = 98 - 105degF
5.11.* (Artix or Zen) = 138-142degF.

Running the cli below on either kernel 5.10 or 5.11 shows no differences at all.
Code: [Select]
sudo watch -n 0.5  cat /sys/kernel/debug/dri/0/amdgpu_pm_info

Doesn't matter if compositing is on or off. I have a manually launched shell script that controls the AMDGPU fan by launching a python daemon. I don't usually launch it unless I'm done working for the day and am going to game. If I fire that script - the GPU temps normalize - and remain normal throughout running any of the games.

Skyrim and Fallout-4 run better under 5.10.* - Artix or Zen. Better as in 5.10.* is clean, 5.11.* has some artifacts left over - so I'm suspicious about the massive code dump for AMDGPU that AMD put in the 5.11 kernel series as being the culprit.

I find nothing in any google/duckduckgo/bing/qwant searching I do.

Again - lets not have anyone burn any time on this - but if anyone has any insights on this...I would certainly appreciate enlightening me.

The bandaid is to launch the python daemon at login, but that's not what I am after.... I'm looking for the underlying reason as to why....

Behavior remains for 5.11.1 thru -.5 revisions (.5 from gremlins repo at this point in time).


Thanks!


Dave



Re: Strange Observation: kernel 5.11.* + AMDGPU

Reply #1 –
Hi,
It's interesting. Would like to find out whether the processor is warmer in fact, or whether it was some code adjustment. Maybe the system was previously received of other (false) information. Or is it a error or bug.

In every new kernel, there is a lot of changes of AMDGPU, so I don't wonder too much. Maybe you can test the kernel 5.12-RC2 too, what it does.

Give here a link to your bug report in bugzilla too.

Re: Strange Observation: kernel 5.11.* + AMDGPU

Reply #2 –
The cause has been found!

I did a bunch of reboots back and forth between kernels and found what was going on. The new AMDGPU code in 5.11 is defaulting the GPU fan significantly slower then the 5.10 AMDGPU code. Thus - the card slowly rises to ~ 140degF.

Simple adjustment to the system and it's resolved.

I used this package from the AUR:
https://aur.archlinux.org/packages/amdfan/

Then followed the directions on the git site:
https://github.com/mcgillij/amdfan

Tweaked the fan curve to my needs in the file " amdfan.yml", put that file in "/etc". Wrote a bash script to launch this app in the daemon mode, added it to xfce's startup settings.....

And all is well.


Dave