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

Appearently no effect #1

Open
MadByteDE opened this issue Mar 11, 2023 · 2 comments
Open

Appearently no effect #1

MadByteDE opened this issue Mar 11, 2023 · 2 comments

Comments

@MadByteDE
Copy link

MadByteDE commented Mar 11, 2023

Hey, thanks for doing this!

Unfortunately, the script seems to have no effect for me.

On Windows the PBO Tuner seems to reduce temps by ~5-10degrees on idle and never spikes higher then ~76degrees.
Running your script does not result in any changes in temps.

Any idea what it could be?

The readback shows -30.

System info gist

Thanks in advance!

@svenlange2
Copy link
Owner

I can only guess. You are comparing very different scenarios. As windows is highly optimized and quaranteed to idle at lowest possible power consumption, I cannot say anthing about Manjaro Linux or your configuration. Maybe your linux box was not using c-states at all or has some background process stressing the CPU, hence you did not see drop in idle. Likewise you have to understand that temperature drop at high load is only visble when you stress all 16 virtual core on parallel - at lower utilization the cores can still run at pretty high speed and undervolting will not push it further (I tested cinebench with 8 cores and there was no large effect with undervolting, but 16 core test showed clear difference in performance as temps were lower).

In my own case, undervolting at -30 i and running Windows in KVM and Cinebench inside the Virtualized Windows, I can see identical drop in temperature as on bare metal Windows (about 75C max vs 90C without undervolting). About Linux: I use Proxmox virtualizer which is basically a Debian.

A good indicator of the undervotling working is when you monitor all core frequencies during a stress test. Usually single core threads can go up to 4450 MHz on my CPU. Now when I stress all cores they go to 4200 or below and with undervotling I can get them all uo to 4400. Thats a little better indicator than checking the temperature that can lag.

@chirvo
Copy link

chirvo commented Apr 25, 2023

Just to let you guys know that I compiled PyTorch twice, before and after running the script (with params -c 8 -o -30) and noted a difference of ~5 degrees less, tops of 85C (avg 84C plus throttling to 4.20GHz) and 80C (avg 79C almost no throttling, 4.40GHz-4.45GHz). CPU governor was set to performance.

CPU: 8-core AMD Ryzen 7 5800X3D (-MT MCP-) speed/min/max: 3603/2200/4549 MHz
Kernel: 6.2.10-060210-generic x86_64
Mem: 64207.8 MiB

EDIT: Added sys specs

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

3 participants