Re: CPU temp?
On Sun, Jun 01, 2008 at 09:23:42PM +0300, Dotan Cohen wrote:
>
> An average of what? According to sensors, the machine only has temp
Tcore1+Tcore2/2 but ...
> sensors on the CPU dies, not on the hard drive or memory or whatever.
> And the apci and /proc temps are not the average of the core temps.
... see the point about different algo/params.
That's likely also different from lm-sensors, which knows of off-chip sensors
(beneath the CPU).
> Really? I've melted older (pentium 3,4) CPUs at around 80, probably
> even less. I don't know how far the sensor on the P4 is from the die,
silicon power devices can have Tlimit 175 degC, nanometer ICs are more
delicate of course, and each CPU family from each vendor comes with its own
in-spec and phy.damage T limits. Hopely the chip 'knows' its own limits,
and it shutdowns itself before melting. If its internal table says 100 degC,
well, that's likely so (though CPU vendor might keep the real value
undisclosed, for the usual reasons).
But if you're curious, check the very datasheet from the URL posted by others.
Yes, old CPU could melt rel. easily if allowd to, but not @T you mentioned,
you read the T from off-chip sensors, (to some extent close to case T),
which means likely the silicon itself was well above 100 degC, with hot spots
> 150 degC.
(Recent) CPUs check their T from on-chip sensor (basically, a diode), which
may be readable by some means (eg via MSR) by eg a program the same CPU is
running.
That's what the module coretemp does.
> but it can't be _that_ far. AMD's I've seen get that hot and live
afaikt, even old AMD's had on-chip arrangements for internal cut off
@T limit. I couldn't burn a K6 even leaving it without heatsink.
OTOH the electronics surrounding the CPU fail{s,ed} frequently, so there's
a lot of spare old CPU around ;).
--
paolo
Reply to: