Hello, Some days ago, I've stated, that the formula "temp2 (@*30/43)+25, (@-25)*43/30" works better than "temp2 @*2, @/2". Today I continued experiments with my machine, and I found that still there are systematical discrepancies between CPU temperature measured from the Celeron diode, and temperature of thermistor probe put near the CPU core, under the heatsink - especially for idle state (temperatures around 30?C). To maximize temperature changes between idle and full-load states I overclocked my iCII 600 to 900 MHz, and raised V_core to 2.0 Volts (well, I know ... but these CPU's are quite cheap nowadays :-) ). The chassis was closed and the temperature inside was about 29?C. Full load was done by using burnP6 from CPUburn suite, half-load was done with same tool, but with the help of the following script: del=50000 while [ true ] ; do kill -SIGSTOP `pidof burnP6` ; usleep $del kill -SIGCONT `pidof burnP6` ; usleep $del done 2> /dev/null I found, that the best formula for temp3, which makes agreement between it and temp3 is "temp2 @+19, @-19". Here you can see the results: http://www.camk.edu.pl/~gawrysz/cuv4x.png Disclaimer ;-), and some comments I tried to put the thermistor as close to the CPU die as possible, but I'm not sure how close it is. It can introduce errors proportional to (T_CPU-T_ambient), so It's still possible that proper temp2 scaling needs a multiplier slightly greater than 1, but I'm sure, that offset as large as 25?C isn't good, but the "round" value of 20?C is very probable. The readouts from the CPU diode are very noisy. I observed them in BIOS - they were noisy too (up to 7?C). This is strange; I haven't seen such behavior on my mother's machine build of Soyo SY 6BA+III (BX, w83782d) and iCII 566 at 810MHz. But the voltage readouts are very noisy too, and I've checked this power supply powering another computer, and got much cleaner picture, so I suppose that the noise comes from as99127f and its interface. Validity of the formula "temp2 @+19, @-19" can be supported by some (rather "hardcore") tests witch CPU fan slowed down or even unplugged, but this needs to find FSB and V_core at which this CPU can run stable at equilibrium temperature about 70?C or a little bit more but currently I don't have time for such educational games ;-) I'd better check validity of temperature measurements for some another machines I'm administrating (A900 on Asus A7V, A1333 on A7V133, and Athlons MP on Tyan Tiger MP and MPX (fortunately Tyan uses Winbond's sensor chips)), and ensure that daemons can shutdown them quick enough in case of CPU fan failure ... -- Regards, Artur Gawryszczak
