[rescue] This Just In: HP to buy Compaq
Scott Newell
rescue at sunhelp.org
Sun Sep 16 15:32:10 CDT 2001
>The PPro and P-II sensors are purely thermal cutout devices and they are
Nope. As I explained earlier the P2-350 and later chips _do_ have an
on-chip thermal diode (in addition to the THERMTRIP 135=B0C thermal cutout
switch). Both ends of the diode are brought out to pins, so you can get an
accurate measurement of the die temperature (with external circuitry) by
measuring the relationship between current through the diode and voltage
across the diode.
>specifically tuned well above the recommended operating range (130=B0C for
>the PPro). they cause the processor to instantly shut down. they are
>useless as management devices.
Here's how I got interested in CPU thermal stuff: I once had some NatSemi
samples of little silicon temp sensors in sot-23. I drilled a hole in the
heatsink and epoxied one next to a P-133, wiring the analog temperature
output to a really neato 4 inch analog Simpson meter movement. I even
printed out a meter face calibrated in =B0C. I remember noticing that dos,
win95, and win98 would run the chip really hot, whereas NT and Linux ran
much closer to room temperature with light processing load (as did some dos
code that ran constant halt instructions). I had a small fan (probably too
small), so the thermal time constant was short enough that you could see
the needle swing when the processing load changed. It was all great fun.
I also learned that the BSOD of NT4 would cook the chip--I guess at the end
of the line, it wasn't executing a halt instruction but instead going into
an infinite loop. That still seems dumb to me. I even tried to get
Sysinternals guys fired up about this, but they saw it as a commentary more
on my heatsink and thermal environment than a flaw in NT.
A friend got interested in building some kind of interface, perhaps to a
serial port, to monitor the CPU temp. He wanted to write a driver or mod
the kernel of FreeBSD to keep the chip from overheating at very high
loading--our best guess was to add a high priority task that would halt the
CPU until the next tick. Never got past the 'wouldn't it be neat' stage,
however.
newell
More information about the rescue
mailing list