[lm-sensors] Wrong temperatures reported for Core2Duo CPU in Intel DG965WH motherboard

Jean Delvare khali at linux-fr.org
Wed Mar 10 19:10:03 CET 2010


On Wed, 10 Mar 2010 08:41:52 -0800, lmsensorslist at nodivisions.com wrote:
> Thanks for your reply.
> 
> On 03/10/2010 10:07 AM, Jean Delvare wrote:
> >
> > The value is reported by the CPU itself. The coretemp kernel driver is
> > just passing it through. While the value is reported relative to the
> > critical limit, and we have had this critical limit wrong for some
> > models in the past, this is irrelevant here: the fact is that your CPU
> 
> Can you explain this a bit more?  You say coretemp is just passing it
> through, which to me implies no calculation nor conversion of any kind. 
> But that's not really the case, is it?  The CPU doesn't literally say "79"
> to mean "79 degress Celsius" right?

You're right, it doesn't say 79. It says 21, meaning "21 degrees
Celsius below the critical temperature limit". As we assume this limit
to be 100°C for your CPU model, we end up with 79°C.

> > runs over its high limit when loaded. I've never seen any Intel Core2
> > CPU reaching such a high temperature. This can mean only 2 things:
> > * You happen to have a CPU with defective thermal sensors. The fact
> >   that both cores agree makes me skeptical, but then again, as I have
> >   no idea how the internal sensors could go wrong, it might as well
> >   happen to both sensors at once.
> > * Your heatsink and fan are not doing their job properly. This is my
> >   favorite theory at this point.
> 
> It's certainly possible that I've messed that up somehow.  However, I've
> installed many dozens of CPUs and heatsinks as a system builder over the
> years, including 2 different ones in this particular system -- and,
> there's basically nothing to it with Socket 775: you just sit the heatsink
> on the CPU and push down the 4 pins.  I'm using the pre-applied thermal
> compound on the Rosewill heatsink, but with the Intel heatsink that I was
> originally using, I used Arctic Silver.  In both cases these high
> temperatures were present.
> 
> > I would not be surprised that the CPU survives 3.5 years at these
> > temperatures. You are still below the critical limit.
> 
> So regularly running in the 80-90 degree Celsius range won't necessarily
> damage the chip?  I see what you mean that it's below the critical limit
> of 100C, but it still seems like 80-90C might hurt it too.

Well, that's pretty high, yes. I wouldn't like it for my own machine.
And probably your CPU won't live 15 years at this temperature. And if
your fan breaks, your machine will die immediately. But... again, as
long as you're below the critical limit, it should be mostly OK.

> > Hard disk drive temperature is a totally different thing. Assuming
> > there is some space between the CPU and the HDD (and there always is on
> > a desktop board) there is no reason for these temperature to match.
> 
> I know they are totally separate sensors, I just wanted to point out that
> for example it's not the case that the whole PC is running hot, or just in
> a really hot room, etc.  It's just these numbers for the CPU cores that
> are high.
> 
> > The QST SDK might let us provide support for motherboard sensors, but
> > this will not change the values reported by the coretemp driver. Best
> > you can hope is extra temperature sensors for comparison purpose.
> 
> I'm hoping for that along with maybe fan RPM readouts, and ideally being
> able to control the fan RPMs.
> 
> > What you can do is check the temperatures reported by the BIOS. It's
> > difficult to know where the BIOS gets the values from, though...
> 
> OK, I just tested this, by running the "stress" command for a few minutes,
> until the coretemp values were 91C and 92C.  I then immediately shut down
> and went into the BIOS.  Here's the BIOS numbers:
> 
> 68C CPU Die/Package Temperature
> 54C Motherboard Temperature
> 44C Motherboard Temperature
> 82C ICH Temperature
> 58C MCH Temperature
> 
> And here's the same BIOS numbers after sitting there in the BIOS for 10
> minutes:
> 
> 63C CPU Die/Package Temperature
> 50C Motherboard Temperature
> 44C Motherboard Temperature
> 83C ICH Temperature
> 54C MCH Temperature
> 
> I guess the die temp is the one that corresponds to the CPU?

Yes, it probably comes from the CPU internal sensors, although only
disassembling the BIOS could tell for sure.

> But why wouldn't it just show the 2 core temps?

Probably because it would need extra code and the author decided
wasn't worth it. The BIOS code doesn't need to be multi-core-aware.
There's little point in monitoring the temperature of a core you're not
even using.

> Do you think it's possible that it really dissipated 20+ degrees during
> the ~1 minute it took me to shut down and get into the BIOS?

It's not impossible. Especially as you said that the heatsink wasn't
that hot to the touch despite the high temperature value, this suggests
that your heatsink and fan are able to extract the extra temperature.

I admit it's a bit of a paradox though, as my first guess was that the
heatsink and fan was insufficient. Maybe you just happen to have a CPU
mode which heats a lot.

You might get a better view by comparing Linux idle with BIOS idle.
Some BIOS are in a tight loop so they are better compared to a
moderately loaded Linux, but the numbers above suggest this isn't your
case (the CPU temperature in the BIOS was decreasing over time.)

Again, one possibility is that we got the critical temperature limit
wrong for your CPU and it is lower than the driver thinks. But that
doesn't change the fact that you're running 21 degrees below the limit,
which is the smaller merging I've ever seen for this CPU model.

> > Fan control is not possible with only the coretemp driver, sorry. You
> > might look in the BIOS options for an automatic fan speed control
> > option, but given how hot your CPU already is, I wouldn't dare enabling
> > it.
> 
> It's already set that way.  The CPU fan is silent when I first boot the
> PC, and the idle core temp is ~75C.  But as soon as I do anything, the fan
> kicks on quietly; then after 10-20 minutes of uptime and regular work,
> it's running at or near full throttle.

Not surprising, given the temperature values you're shown.

-- 
Jean Delvare
http://khali.linux-fr.org/wishlist.html




More information about the lm-sensors mailing list