[lm-sensors] Interpreting lm-sensors's output
khali at linux-fr.org
Tue Mar 20 08:47:07 CET 2012
On Mon, 19 Mar 2012 23:01:48 -0400, Stefan Monnier wrote:
> >> k10temp-pci-00c3
> >> Adapter: PCI adapter
> >> temp1: +51.5°C (high = +70.0°C)
> >> (crit = +100.0°C, hyst = +97.0°C)
> >> OT1H the "high=70" and "crit=100" would seem to give me the answer, but
> >> OTOH I don't know whether I can trust them (actually, I hope some of
> >> the output is "wrong" since the sensor readings go up routinely to 80°
> >> during long compilations).
> > They are correct and you should trust them. If you reach 80°C then you
> Just one more question about these numbers: would the same numbers
> appear under Windows (everything else being "equal")?
This is impossible to say, I'm afraid.
First, there are various ways to save power in modern systems, via
frequency scaling, undervolting when idle, P states, C states, etc.
While I believe Linux is doing reasonably well these days in this area
as far as CPU is concerned, there can always be specific cases where it
Secondly, the monitoring values returned depend on the tool used to
gather, interpret and present them to the user. You've seen yourself
that the k10temp, radeon and it87 drivers report different
temperatures, so you have to know which one the other side is using
when comparing. Also note that some tools can apply arbitrary offsets
to the temperature values they report, to compensate for a number of
things (like thermal sensor imperfection or distance between sensor and
monitored item.) This is something we can in general not do under Linux
as we do not have the information.
So, in the general case, yes, comparing numbers between Linux and
Windows makes sense, but you have to be cautious as there are
exceptions. The only way to make sure would be to actually run Windows
on your board.
Things are even more complex in your case because the GPU and CPU are
grouped in a single package on your system, so they influence each
other. If you're using the open-source radeon driver, I invite you to
play with the power settings:
For example, forcing the profile to "mid" during CPU-intensive tasks
may help. dynpm method may be even better but I never dared to try it.
Also note that the binary driver (fglrx) is probably better in terms of
power management, and would be closer to what Windows has, so this may
be another thing to try.
> PS: I'm trying to argue with the company (Zotac) that my board has thermal
> problems, but they suck (i.e. they dismiss GNU/Linux users), so I need
> to be sure that the numbers they give me (which they get under Windows
> from I don't know which tool) are comparable to the ones I see.
This is strange, I was in contact with a Zotac engineer in late 2009 /
early 2010 and back then they seemed to care about Linux at least to
some extent, in particular when it came to thermal monitoring.
More information about the lm-sensors