Hotter GPU lately

Lately my AMD Radeon 7800XT has become hotter in Bazzite it hovers around 64 degrees celsius in idle on memory and about 58 degrees on junction, and sometimes more so that the fans go crazy to cool it down. This problem does not exist in Windows. There I have normal temperatures, about 45 degrees idle. And fan curves works.

I have installed CoreCtrl, via ostree command, but the option to make fan curves is greyed out so there is no way to control fans (same thing in Lact). Fans are either not running until temperature hits 67 degrees and then on max speed so it is loud.

Is this a AMDGPU driver issue? or is it because sensor nct6687-isa-0a20 (that I used to control the fans with) has been removed from the kernel?

Hi @hakinger,

nct6687 sounds like a motherboard sensor chip so I don’t think it has anything to do with your graphic card. Graphic card has it’s own sensor and fan control chip. Some standard chips are supported by the graphic driver (AMDGPU) but a card vendor can decide to use some custom or exotic sensor chip which is only supported by vendor application (usually combines monitoring and overclocking features).

You’ll have to check for your particular model and vendor of graphic card what sensor chip is used, either trough Googling or by visual inspection (cooler removed and all). Then you will have a better idea if your sensor chip is supported or not.

Sometimes a sensor chip is supported but temperature readouts are wrong. This happens because temperature is measured as a voltage on a sensor output. This voltage is then used in a specific formula to compute the temperature. The formula is often not known and the exact formula is only known by the vendor. Formulas can vary wildly among vendors and models of graphic cards. Some formulas are reverse engineered, for example by comparing temperature readouts in vendor supplied application for various values of raw voltage readouts or by reverse engineering vendor applications. But it is not viable to reverse engineer every model of every vendor.

It is unfortunate that vendors are not willing to disclose this kind of information. It would make things much easier.

Last but not least, some GPUs have builtin sensors which makes them universal/standard among vendors and models but a vendor can still decide to put additional sensor and fan control chip. What happens then is that vendor application will hide values from builtin sensors and show only values from additional sensor chip. On the other side, when you go to Linux and use lm-sensors to display sensor values, it will show values from builtin sensors which can show considerably different values that additional sensor chip.

Complicated :worried:

1 Like