Question CPU temperature sensors defective?

[creative name]

Hello,

I am not doing anything demanding on my machine, just office work, and I let it run through the night. This night it had crashed, displays black, not waking up, had to be restarted. One hour later the image "froze" - another restart. I checked CPU temperatures with Core Temp. It seemed to be normal in terms of how high they are, but fluctuate wildly: Sometimes without a change in the fan noise or load on the CPU, they jump from 24°C to 34°C within a second, and a few seconds later are back down to 24°C. What is particularly strange, is that they can reach as low as 15°C, which is below my room temperature at 21°C.
So I am wondering, if the temperature readings are wrong and thus the fans not working when they should be working, it could explain the crashes.
My CPU is a (rather old) AMD A10-5700 (APU) on an FM2 socket. So it is not easily replaced, one would have to replace the mainboard and memory, too and probably get a new windows and re-install everything, which would take a day or so. Thus I don't want to fix it by replacing the system. The PC continues to be good enough for office work.

Edit: I attached a recording of Core Temp. The temperature reading is cut off at the bottom. I think it is still readable, but I could not see how to fix it (Windows Game bar for recording).


Thank you very much for any ideas.

 

[Lafong]

First thing I'd do is use another temp sensing program to see if you get the same results. I use HWInfo64, but there are others. The lower than ambient doesn't inspire confidence.

 

[creative name]

Thank you. HWiNFO64 has a stable CPU temperature of 40-45°C (more plausible than 24°C) and their CPU DTS reading coincides with what Core Temp shows as the CPU temperature. CPU DTS, according to HWiNFO64, is provided by a temperature sensor on the mainboard. Since these temperatures fluctuate between 11°C and 38°C, I think it is safe to say this sensor is not working as it should. However, the CPU fan is most probably not connected to these faulty readings.
The CPU Package temperature looks to be on the high side. The CPU (Tctl) reading has a too wide range and, again, temperatures below ambient (see screenshot). CPU Package (TSI) temperature is exactly 49°C above the CPU (Tctl) and is moving around wildly in tandem with it. HWinfo says these readings are not necessarily reliable. The one reading that should be reliable is what they label "CPU" at 42°C, which is provided by a diode underneath the CPU. And these temperatures look good and do not seem to explain the crashes.