macmon icon indicating copy to clipboard operation
macmon copied to clipboard

M4 :: CPU/GPU temperatures are wrong

Open iDoka opened this issue 1 year ago • 11 comments

Thanks for awesome utility!

On mac mini m4 I see strange CPU/GPU temperature. My first thougth was that temps are mixed: CPU <-> GPU (because GPU on heavy load but only CPU temp is quite high). But 9.8C suspicious low for room environment (25C as usual).

Screenshot at 2024-12-08 22-09-14

macmon debug > mac-mini-m4-macmon.log

mac-mini-m4-macmon.log

iDoka avatar Dec 08 '24 18:12 iDoka

Hi, @iDoka. Apple has changed the keys that hold the temperature values. As far as I understand, it is currently unknown which keys are responsible for what.

There's a thread here where community are trying to figure it out: https://github.com/exelban/stats/issues/2249

I'll leave issues open, maybe I'll find something.

Also issues in different apps:

  • https://github.com/crystalidea/macs-fan-control/issues/791

vladkens avatar Dec 16 '24 19:12 vladkens

Hi @vladkens Thanks for your comment. If you need to try something on the M4 HW env to fix temperature issue, just ping me in this issue.

iDoka avatar Dec 16 '24 22:12 iDoka

I have a M2 Ultra Studio and I'm always getting 0 CPU power reading. Should I open another issue?

i0ntempest avatar Dec 17 '24 14:12 i0ntempest

Hi, @i0ntempest. Yes, open please new issue and attach macmon debug output to it.

vladkens avatar Dec 18 '24 07:12 vladkens

Hi @iDoka.

If you have a change, can you test this apps:

  • https://www.tunabellysoftware.com/tgpro/
  • https://crystalidea.com/macs-fan-control (https://github.com/crystalidea/macs-fan-control/issues/786).

They both reported that they support the new socket.


Also your terminal window looks small, so the GPU temp value may be clipped. Given that you have a GPU utilisation of 100%, the real displayed value is 79 or 89.

But it's still not clear to me if you have any GPU value at all. In the log you provided, no Tg values in the SMC list, so the temp value should be 0.

I have released version v0.5.1 with the full list of T keys. Can you please also attach new log file.

vladkens avatar Dec 22 '24 14:12 vladkens

Hi @vladkens thanks for update. I will do macmon debug > mac-mini-m4-macmon.log after updating macmon in brew. Now only v0.5.0 is available. BTW, you are right - my window was not so wide enough and GPU temp was clipped (my bad!).

https://www.tunabellysoftware.com/tgpro/ https://crystalidea.com/macs-fan-control

Thanks for sharing links for alternatives, to be honest: my mac mini m4 has headless setup and I have only SSH connection to its, so macmon and asitop only suitable for this case.

iDoka avatar Dec 22 '24 15:12 iDoka

v0.5.1: /opt/homebrew/bin/macmon debug > mac-mini-m4-macmon.log mac-mini-m4-macmon.log

iDoka avatar Dec 22 '24 19:12 iDoka

A couple of interesting things:

  1. GPU temp is gone (just disappeared).
  2. RAM usage differs than reported asitop (not critical for me, but interesting)

Screenshot at 2024-12-22 23-37-22

Screenshot at 2024-12-22 23-37-54

iDoka avatar Dec 22 '24 19:12 iDoka

UPD: v0.5.1 shows GPU temp seems correct (no temp shows for GPU when it in idle/stdby):

Screenshot at 2024-12-23 00-11-51

iDoka avatar Dec 22 '24 20:12 iDoka

  1. GPU temp is gone (just disappeared).

Yes, apparently this is a new behaviour on new chips that CPU / GPU sensors don't work when there is no load on the dice. TgPro article for reference: https://www.tunabellysoftware.com/blog/files/tg-pro-apple-silicon-m3-series-support.html

  1. RAM usage differs than reported asitop (not critical for me, but interesting)

RAM can be calculated in different way. On OS level RAM calculated in pages of fixed size and this pages can be taken, free and many more statuses. I used some formulas from stack overflow to calculated used bytes, but also adopt it to be close to Activity Monitor values. Probably asitop (powermetrics) uses some different approach (funny, while both is apple software)


About issue in general I added smooth function for temp sensors. I can't test it on M4 chip directly, but tried to emulate behaviour when sensor have no data and it looks better now.

vladkens avatar Feb 25 '25 02:02 vladkens

I think that the issue you referenced was resolved? https://github.com/exelban/stats/issues/2249

damageboy avatar May 17 '25 08:05 damageboy