I’m running some Spark jobs on my Framework Desktop (Ubuntu 25.10) and I’m getting the temps below (this is obviously a snapshot). I see it spiked up to 100 C, but those are fairly short lived. The temps typically peak for longer periods between 90 to 97 C and will sometimes dip below that as well.
What are the safe operating temps for this desktop? Is it safe to average above 90 C and peak at 100 C for long periods?
The temps generally look good except for the ‘temp4‘ temperature and the ‘CPU‘ temperature. Does the Framework Desktop support undervoting the CPU?
I also have no Idea what the elusive ‘temp4‘ represents (maybe memory junction temp?) and I’m not familiar with psensor. Does it work kinda like xsensors?
What are your power settings? You can check it eg with s-tui if you run it with sudo.
The tool Ryzenadj can do all sorts of wonderful stuff, undervolt the CPU, set fully custom power limits and even reduce Tjmax to a custom value. Be aware that you should only change any of those settings if you know what you are doing. In general, especially avoid increasing power or temperature limits beyond stock settings. I did some undervolting initially, and it looked good but after a while I got weird system crashes, which disappeared when undoing the undervolting. So I just did no bother to figure out a milder undervolting and kept voltage stock since.
On my system I did set tjmax to 88°C (instead of the stock 100°C). While the APU should be able to handle 100°C. I just feel like it is less stress on the material when not going all the way up. It also works very well. No impact on performance as long as it stays below and thermal throttling to keep temps at max 88°C. Performance loss is minimal in my case.
100 °C for long periods is OK for the CPU (assuming throttling is acceptable for your application), but it might be bad for the capacitors on the mainboard in the long run.
according to the motherboard sensors, everything looks fine; <50°C on ambient sensor is very chill
My use case is, let’s say, a bit unique. As I have a fully custom passive cooling system, during continuous full loads (for modern games), things get a bit crispy. I can only guess capacitor temperatures but given that the heat sink is heating up to 65°C and the backside of the board having a similar temperature I would gess the capacitor temperature is close to that as well. The sensor reading would be a few °C lower than that. I am aware that this could shorten the capacitor life span but these temperatures occur only during gaming, not other use cases, where we are talking rather about 30°C. It is a risk I am willing to take, for science.
That is with my thermal throttling at 88°C for the CPU (even though games do not necessarily need any throttling to not exceed 88°C at the CPU. This is primarily required during full single or multi core CPU loads but it can’t be a lot, as performance does not take much of a hit)
I get the vibe 
I have a similar kind of setup as well (not a Framework). Passive cooling is cool, but I eventually stopped chasing it too hard — mainly because it makes coil whine and other VRM noises much more noticeable.
My view is pretty straightforward: I’m basically using the chip, and since I have the option, I don’t really mind letting it run hot under load. Running it like a space heater is an acceptable trade-off for me.
Long term, I’d actually prefer to move this steel box somewhere isolated — a closet or something like that, far away. Also, I find phase-change cooling quite appealing — “GHz that are not for sale” and all that.
I don’t have a FW Desktop, but I have a FW 16 and I run Linux mint 22.2 and I use the GNOME power daemon to set the CPU to power efficiency and for the GPU, I use LACT to set the max power rate to 120W and bootup_default when I wanna play RDR2 or Hogwarts Legacy and set it to power_efficiency profile when I’m not doing much (YT, Work, school, etc.) It cuts the average wattage to a little below 20W which is higher than what it was before I installed the graphics module
