Odroid H4 overheating with Trixie

What happened:

  • Upgraded DietPi to Trixie using the script on my Odroid H4 (passive cooling)
  • Got overheating issues
  • Stopped all services: Still overheating
  • Installed a fresh image of DietPi Trixie: Still overheating issues!
  • Iinstalled a fresh Image of Dietpi Bookworm: No overheating.

Here is a Table with details:

Install Version htop load temps
services running, before upgrade Bookworm unknown ~ 55 °C
services running, after upgrade Trixie 0.20 70 - 72 °C (thermal throttling?)
services stopped Trixie 0.00-0.01 60 - 63 °C
Fresh default installation* Trixie 0.00-0.01 60 - 63 °C
Fresh default installation* Bookworm 0.00 43 - 45 °C

*Installed NativePC x86_64, docker+docker-compose selected to install. Governor set to powersave with same frequency limit on all installations. Everyting else was left to default settings.

Required Information

  • Architecture | amd64
  • SBC model | Native PC (x86_64) o
  • SD card used | emmc 64gb

Expected behaviour

Trixie cpu temp should be ~20 °C lower

Extra details

  • temp reading seems to be correct, heatsink is hot to touch when running trixie
  • i didn’t see anything weird going on in htop, but I have no experience with htop or linux performance profiling in general
  • i’m running bookworm again for now

I might install vanilla debian trixie just to check if it is an upstream debian issue. I will be using Bookworm as a workaround, still posting this to see if anyone got similair problems.

That is a good idea, since it uses the same kernel.

Do you see anything in htop causing unusual high CPU usage? In case also unhide kernel threads in its options. And check

dmesg -l 0,1,2,3

for kernel errors.

Update:
I Installed current debian netinstall and got the same issues (66 °C), so yes, debian is causing this.

And check dmesg -l 0,1,2,3 for kernel errors.

Good Call! There was indeed an error. Sadly, I didn’t note what exactly it was, but i manged to fix it by installing these (graphics?) drivers.

Install: firmware-nvidia-graphics:amd64 (20250410-2, automatic), firmware-mediatek:amd64 (20250410-2, automatic), firmware-misc-nonfree:amd64 (20250410-2), firmware-intel-misc:amd64 (20250410-2, automatic), firmware-intel-graphics:amd64 (20250410-2, automatic)

Since then, the temps are good again (48 °C)!

I have no idea what was going on. Why is was this suddenly missing although it worked on bookworm?

I will try to install dietpi (trixie) and check if this can be also fixed in the same manner.

just tried it again with dietpi, same problem:

root@DietPi:~# dmesg -l 0,1,2,3
[    3.816973] i915 0000:00:02.0: [drm] *ERROR* GT0: GuC firmware i915/tgl_guc_70.bin: fetch failed -ENOENT
[    3.818042] i915 0000:00:02.0: [drm] *ERROR* GT0: GuC initialization failed -ENOENT
[    3.818051] i915 0000:00:02.0: [drm] *ERROR* GT0: Enabling uc failed (-5)
[    3.818056] i915 0000:00:02.0: [drm] *ERROR* GT0: Failed to initialize GPU, declaring it wedged!

the fix was apt install firmware-misc-nonfree and update initramfs

Should this not be installed by default? Firmware is quite important!

nonfree is never installed by default. Maybe on Ubuntu it is :smiley:
see https://wiki.debian.org/Firmware#What_is_firmware.3F

I never understood this purism. It leads to users just being confused why stuff doesn’t work.

This is not about purism, in this case it is nVidias fault.
Debian has guide lines how to handle software. nVidia does only provide propiertry drivers/firmware, so they are not included in the default repos, because nobody can tell what they do under the hood, they are closed source software.
E.g. AMD on the other hand has open source drivers, they are included in the default repos.

But there are distros which include non-FOSS software by default, like Ubuntu or Linux Mint or MS Windows. Maybe this is a better option for you then?

But what you just said means that it kinda is about purism: it is Debian’s community that sets the Debian guidelines.

I don’t want to defend nVidia or whoever, and it’s good to have specific guidelines, but if I was installing Linux for the first time, I would be super confused about why my hardware wouldn’t work. It is what deters people from switching to Linux. You just want to run a game or watch a film and suddenly you learn (or not) that there is a litany of commands you need to run before you are able to. The promise of Linux is to have drivers preinstalled; in theory they could be, but aren’t because developers can’t have a look (the user doesn’t care).

nVidias fault

For the record: I don’t even have a discreete gpu, the issue seem to be some Intel driver.

The issue is solved now, but as a somewhat new user I would prefer an opt-out for unfree drivers instead of an opt-in.

Thinking about in, even an opt-in would be okay. Just a checkbox during install would solve this kind of issues.

Graphics drivers and firmware can be installed with dietpi-config display options as well. But in case of Nvidia it installs and compiles the Nvidia driver via DKMS, probably unnecessary for server systems without much GUI usage, desktop or such.

Weird that the missing firmware is causing idle CPU load, unless some X or GUI session is running.

On Bookworm this was not an issue since the firmware-misc-nonfree package contained the firmware, which was split into firmware-nvidia-graphics with Trixie. It found that actually a good thing, since that firmware is not needed for any other graphics vendor. But if that causes a high idle CPU load even on a fresh Debian/DietPi, we might reconsider: Maybe we can check for those errors or Nvidia graphics card and install the firmware automatically during first run setup, the same way we install Intel or AMD microcode packages if respective CPU were found :thinking:.

But actually this sounds more like something to fix/enhance in the respective kernel module.