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

Jean Delvare khali at linux-fr.org
Fri Mar 12 11:20:45 CET 2010

Hi Anthony,

On Thu, 11 Mar 2010 12:38:05 -0800, lmsensorslist at nodivisions.com wrote:
> > Today I installed Windows XP on this same system, to see what it would
> > report for the core temperatures.  I installed 3 apps for this: SpeedFan,
> > Core Temp, and Real Temp.  Here's what they reported while idle and after
> > 10 minutes of high load (all 3 temperature apps reporting simultaneously):
> >
> > SpeedFan: idle 49C, loaded 70C
> > Core Temp: idle 54C, loaded 75C
> > Real Temp: idle 59C, loaded 80C
> >
> > The Core Temp & Real Temp apps explicitly show what they're using for
> > TjMax: 85C and 90C respectively.  And from the idle/load temps it looks
> > like SpeedFan is probably using 80C.  lm-sensors/coretemp are 10 degrees
> > higher still.

Very interesting that apparently nobody agrees on what TjMax is.

> > So which is correct?  Is the TjMax value published by Intel, and/or
> > definitively known to be a certain value?  I searched intel.com and the
> > datasheet for my CPU (the E6600) but didn't see it.  And this tomshardware
> > piece seems to say that Intel deliberately obfuscates this information
> > (read the update at the bottom):
> >
> > http://www.tomshardware.com/news/intel-dts-specs,6517.html

This is our experience as well, we asked Intel repeatedly about the
TjMax values for various CPU models but couldn't get a clear answer.

> > If SpeedFan is correct, I feel much better about my CPU; but if
> > lm-sensors/coretemp is right, I'm more worried.

Unfortunately you will probably never know. The only reliable data point
here is the difference between the operating point and TjMax, not the
absolute temperature values.

> Sorry for the double-post, but I meant to re-iterate just for
> completeness' sake: the idle/loaded temps reported by lm-sensors+coretemp
> under Linux are ~73C/93C, in contrast to the lower temperatures reported
> by the 3 Windows apps mentioned above.

Thinking in relative terms, your system is running under Windows at -31
idle and -10 under heavy load; Linux at -27 idle and -7 under heavy
load. So the bottom line is that your system runs hotter in Linux than
Windows. You should check if you have services running that suck your
CPU, or, if you self-built your kernel, if you missed some CPU-related
options such as CONFIG_CPU_IDLE, ACPI, etc. And also make sure you have
the acpi_cpufreq and cpufreq_conservative drivers loaded and used. The
"powertop" utility might come in handy to track such issues.

The 20°C difference between idle and heavy load seems reasonable, my own
mobile Core Duo CPU has an even larger difference (37°C!). The 7°C
margin is more problematic, but that being said, I just checked mine
and it's running only 12°C below the limit under heavy load... That's
not so different from yours. So maybe these CPUs really run hot and so
be it.

Jean Delvare

More information about the lm-sensors mailing list