lm_sensors on Fit-PC2

[Denis]

Hi prj,
I see now my script reports 51-53 C on fit-PC, yours reports 31 C. I think the real temperature is much closer to 51-53 then to 31. I will check it next week with infrared thermometer.

[prj]

Thanks, that would be great. If you have any info on Tj_max other than what is mentioned in the Intel documentation I would be glad if you could share it.

I agree that 31 C sounds low, but on my unit the case surface temp seem to follow my temp readings quite accurately. Then again... my equipment for measuring is quite crude. If MSR readings are wrong it would be nice to find a way to get them right so it can be fixed in the next kernel release.

[Denis]

Hi all,
Tested with infrared thermometer. Measurement on cpu surface gives about 7 degrees Celsius less then script does. So seems script on the first page of the thread gives correct core temperature.

[prj]

Thanks Denis for taking the time to check. Would you say that the 7 C difference where constant to the actual temperature? Perhaps a Tj max of 100 C is a better match? If this is the case it can be adjusted in the coretemp driver and perhaps backported to Gregs 2.6.32-stable tree. At the moment (from what I can see in the source) we will get a 90 C Tj max and thus an incorrect reading in Ubuntu 10.04.

[PieterPan]

I will crank up both cpus to try and raise the cpu temp above 50, then I can compare both outputs. But I will probably have to wrap it in something as my heatsinks are very effective in keeping the cpu's below 50 (and the heating at home is off, so low ambient temperature). Follow it here: http://home.panman.eu/stats/

Edit: Nope, too cold there. Will try again tonight with the fit-pc wrapped up.

[prj]

PieterPan, nice stats! Is it possible to get the load into the stats as well? I still get very strange readings from the first script. Perhaps my sensors are malfunctioning. At very low load it reports around 60 C (Sensor 0) and at full load it usually goes down 1-2 C or just stays the same. IMHO it seems very random.

[PieterPan]

For me it follows CPU load very well. The big drop you see in MSR temp must be the result of me turning of both 'cat /dev/zero > /dev/null' processes. I can add them, but not right now

[PieterPan]

Ok, it is now wrapped up and warming up nicely. You should be able to see the cpu temp measured from the first script increase soon and hopefully it is a fixed offset, so it will just be a matter of changing the tj_max.

edit: I've added a line for the difference (dark green), it seems to hover around 15 degrees as soon as the compulab script goes above 50. Maybe the TJ_max in this case is 105?

Edit2: Ok I think that is warm enough I've unwrapped it, back to normal. Perhaps this empirical data will help. Very interesting observation is that the MSR values drop much quicker than the compulab script, so I expect them to be different sensors.

[prj]

Since the Compulab sensors don't behave like CPU sensors it might be possible they are located on the US15W. If stressing the GMA500 shows a quick increase in temperature this is likely the case. There is a kernel module (i2c-isch) that claims to support the US15W but I cannot get any readings from it.

[PieterPan]

Well, whatever it is, it seems to work ok when I add 15 to the msr value, which I did. Looks a lot nicer on my plots, and I don't have to worry anyway with these cool temperatures. Scientific, huh?
Thanks!