Re: CPU temp -268 degrees!

On Sat, 10 Nov 2007, in the Usenet newsgroup comp.os.linux.misc, in article
<fh4nrn$4jh$1@xxxxxxxxx>, Andreas Wenzel wrote:

Moe Trin schrieb:

You can't. If you read the specification sheets for the sensors, or
for the CPU, you will find these are only crude estimates. Really.

Now you made him feel bad.

Sorry - it's a fact of life.

I did a short study on temperature reporting systems in PCs - CPU,
motherboard, and SMART drives - and the accuracy figures all suck. Part
of this is due to the method of measuring the temperature, and a larger
part is due to the physical placement of the sensor, and the fact that
the manufacturer can't afford to "calibrate" the measurement system.

CPU - measure voltage drop across a test diode located somewhere in the
middle of the die. Problem: silicon is a lousy conductor of heat, so
the temperature may not be representative of hottest stuff on the chip.
Problem: voltage is a direct function of the temperature measured in
degrees from absolute zero (-273.18C), so a ~3 degree Celsius change is
one percent. Problem: measuring device has a wide input range, so
gain and offset is needed to get usable scale factors - and both gain
and offset are controlled by 5% tolerance components. Problem: this
is consumer gear, and the manufacturer can't afford the time or cost of
parts to "calibrate" the sensor. Intel specifically states that the
operation of the diode is not tested, period.

Motherboard - measure voltage drop across a thermistor pill located
somewhere on the motherboard. Problem: location not specified -
what is it actually measuring? Problem: thermistor pill sensitivity
and setting a function of the mixture of material that makes the pill,
and subject to manufacturing tolerances - WIDE (5 - 10%) tolerances.
Problem: this is consumer gear, and the manufacturer can't afford the
time or cost of parts to "calibrate" the sensor. And, the board
manufacturer is buying the cheapest possible component that seems to
work with an acceptable failure rate.

The 'SMART' drive sensor has the same problem.

It's unlikely that the temperature sensing function is even _tested_ at
all (beyond a "yes, it's indicating something" vs "seems to be dead").
Remember that your PC was probably assembled by untrained chimpanzees
who are paid four bananas a day. The "acceptance test" is that the
system doesn't smoke when powered up, and appears to BIOS self test,
and boot windoze95 without crashing more than usual.

The bottom line is that the reported temperatures are crudely measured,
and are probably quite inaccurate. The best you can do is to make
relative comparisons of the conditions of a particular (serial number)
sensor. If it's been reading $FOO for a year, and now suddenly is
reading %BAR, then something has probably changed. Use a calibrated
thermal sensor to find out what may have happened. The household
thermometer is useless - next time you are at the store, look at them,
and note that they're often indicating different temperatures. The
"mercury column" (which is a glass tube filled with a colored liquid
that expands with temperature) may even have a label loosely glued over
the tube so illustrate what the thing looks like - masking that it's
showing the "wrong" temperature. The electronic versions may not even
have a battery installed, and the display is another label showing what
it should look like. Consumer goods, built as cheaply as possible.

Old guy
.

Relevant Pages