Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

amd_pstate not available. #189

Open
jambokill opened this issue Jan 30, 2025 · 14 comments
Open

amd_pstate not available. #189

jambokill opened this issue Jan 30, 2025 · 14 comments

Comments

@jambokill
Copy link

jambokill commented Jan 30, 2025

Hi,

I just noticed that on the kernel config you prefer to disable amd_pstate and use acpi-cpufreq.
What are the benefits in using acpi-cpufreq instead of amd_pstate.

Is there an easy way for this to be enabled or do I have to recompile the kernel and remove this line CONFIG_CMDLINE="audit=0 intel_pstate=disable amd_pstate=disable"

Thanks,
Jan

@hwsmm
Copy link

hwsmm commented Jan 30, 2025

This commit explains it: 07b176e

@damentz
Copy link
Owner

damentz commented Jan 30, 2025

Thanks @hwsmm, it's common for popular distributions to override the power profile of the kernel to serve laptops better. Unfortunately, that improvement in efficiency leads to a large number of common workloads performing very badly. Specifically, single threaded tasks that don't quite saturate the single CPU they run on.

In Zen Kernel, acpi-cpufreq has been tuned to cause the CPU to enter boost mode far sooner than stock (about 60% CPU per core). And on Liquorix, the performance governor is selected by default, but ondemand is also available if you want to tune your system for lower power.

And you can always add amd_pstate=active to re-enable the frequency governor is you prefer.

@jambokill
Copy link
Author

jambokill commented Jan 30, 2025

I tried appending amd_pstate=active on my grub menu but it doesnt work. So do I need to remove that line and recompile the kernel?

Also let's say I don't go with the amd_pstate route, how would power-profiles-daemon behave with the current driver of acpi-cpufreq if I set the governor to ondemand?

Thank you @damentz @hwsmm for entertaining my question.

Best regards,
Jan

@damentz
Copy link
Owner

damentz commented Jan 30, 2025

That's interesting, I tested the disable + active mode on a local system and it worked. Maybe your system is not compatible with amd_pstate?

I can retest this behavior once I have access to my AMD system in the next day or two.

@Insurgent65
Copy link

In my case adding amd_pstate=active to grub worked, sometimes we forget to sudo update-grub afterwards, just a possibility.

The problem for me is that utilities like cpupower_gui don't work without amd_pstate.

On the other hand this seems to be on the 6.12 kernel because on 6.13 it doesn't seem to be necessary, I'm anxiously waiting for liquorix to update to 6.13.

@jambokill
Copy link
Author

I added amd_pstate=active and updated the grub via sudo update-grub but it still didn't work, I also tried blacklisting acpi-cpufreq but to no avail.

@Insurgent65
Copy link

I added amd_pstate=active and updated the grub via sudo update-grub but it still didn't work, I also tried blacklisting acpi-cpufreq but to no avail.

I recommend you to wait for version 6.13, maybe it will solve itself, I don't think it will take more than a few days.

@jambokill
Copy link
Author

Hi @Insurgent65.

Okay, I will just wait for the new update. More power to the devs and thank you for this wonderful kernel.

Best regards,
Jan

@damentz
Copy link
Owner

damentz commented Feb 1, 2025

I added amd_pstate=active and updated the grub via sudo update-grub but it still didn't work, I also tried blacklisting acpi-cpufreq but to no avail.

I recommend you to wait for version 6.13, maybe it will solve itself, I don't think it will take more than a few days.

The only wait is on an update from Project-C. The developer behind is working on fixing a regression in scheduling plus the sync-up with 6.13. It may be another week or two before Liquorix is on 6.13.

If the dev falls behind too much, I'll take a shot at doing the sync-up myself.

EDIT: Here is the regression I reported if yall are interested: https://gitlab.com/alfredchen/linux-prjc/-/issues/104

@Insurgent65
Copy link

EDIT: Here is the regression I reported if yall are interested: https://gitlab.com/alfredchen/linux-prjc/-/issues/104

you are going to have to make a decision, the deadlines for the fix expired a long time ago, I don't know the extent of the problem either, can we live with this bug?

@carlosfmesilva
Copy link

This exact same issue happens with intel_pstate.

In my machine, if I set intel_pstate=enable, the pstate is enabled but it selects intel_cpufreq as a scaling driver.

If I set intel_pstate=active, the pstate is disabled and it selects acpi-cpufreq as a scaling driver.

Please note that if I install the kernel from the official Debian repo, the intel_pstate=active gives me intel_pstate as a scaling driver.

My installed current kernel version is 6.12.15-1-liquorix-amd64.

@Insurgent65
Copy link

If it helps anyone, I have chosen to stay with the 6.11 version that you can download from here:
https://liquorix.net/ubuntu/

The problem is that it has no maintenance, but it is only while version 6.13 arrives.

@damentz
Copy link
Owner

damentz commented Feb 20, 2025

This exact same issue happens with intel_pstate.

In my machine, if I set intel_pstate=enable, the pstate is enabled but it selects intel_cpufreq as a scaling driver.

If I set intel_pstate=active, the pstate is disabled and it selects acpi-cpufreq as a scaling driver.

Setting intel_pstate=enable simply reverses the intel_pstate=disable flag set earlier. If you want to force active mode, you must pass intel_pstate=enable intel_pstate=active to make sure the disable flag is reverted.

@Insurgent65
Copy link

Liquorix version 6.13 runs smoothly, with a noticeable performance increase.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants