[lm-sensors] A better fan control program

[lm-sensors] A better fan control program

[Paul Crowley]

I know there are several programs out there that adjust your fan speed 
based on temperature readings.  However, I couldn't find one worked well 
on my system, so I wrote my own.   This program takes a very simple 
approach: if the system is too hot, the fans run on maximum power, 
otherwise they run at minimum power.  The reasoning behind this approach 
is explained at length here:

http://www.lshift.net/blog/2006/06/26/thermostat-defeat

and the resulting program, in C, is available here:

http://www.lshift.net/downloads/fankuchen-v0.1.tar.gz

This is now running on my system and I'm very happy with the results.

Should this be the program that ships by default with lm-sensors?  Let 
me know what you think.  Thanks!
-- 
   [][][] Paul Crowley
     [][] LShift Ltd
   []  [] www.lshift.net

[lm-sensors] A better fan control program

[Mark M. Hoffman]

Hi Paul:

* Paul Crowley <paul at lshift.net> [2006-06-28 11:09:48 +0100]:
> I know there are several programs out there that adjust your fan speed 
> based on temperature readings.  However, I couldn't find one worked well 
> on my system, so I wrote my own.   This program takes a very simple 
> approach: if the system is too hot, the fans run on maximum power, 
> otherwise they run at minimum power.  The reasoning behind this approach 
> is explained at length here:
> 
> http://www.lshift.net/blog/2006/06/26/thermostat-defeat

I skimmed through the blog entry and the earlier one which it references.
In your first attempt you reinvented the 'P' part of a PID control loop.
Don't give up! ;)  There's a *ton* of info about PID controllers on the
web.  If you're still interested, you could start here:

http://en.wikipedia.org/wiki/PID_controller

Regards,

-- 
Mark M. Hoffman
mhoffman at lightlink.com

[lm-sensors] A better fan control program

[Paul Crowley]

Mark M. Hoffman wrote:
> I skimmed through the blog entry and the earlier one which it references.
> In your first attempt you reinvented the 'P' part of a PID control loop.
> Don't give up! ;)  There's a *ton* of info about PID controllers on the
> web.  If you're still interested, you could start here:
> 
> http://en.wikipedia.org/wiki/PID_controller

Very interesting stuff, thanks!

My quick reading of this stuff indicates a possible problem.  On my 
system at least, I can only read the temperature in whole degrees; 
lm-sensors returns a number in thousandths of a degree, but sadly the 
last three digits are always zero.  This means that our differentiation 
of the temperature will be very crude, which could make it difficult to 
build a functioning PID controller.

How can I find out the maximum safe operating temperature of my current 
CPU?  Since my first priority is something that works on my current 
system, the only reason for me to pick this up again and explore in more 
detail is if I can afford to slow the fan even when the system is flat 
out, which means I need to know how fast I can run it.

Thanks again!
-- 
   [][][] Paul Crowley
     [][] LShift Ltd
   []  [] www.lshift.net

[lm-sensors] A better fan control program

[Mark M. Hoffman]

Hi Paul:

* Paul Crowley <paul at lshift.net> [2006-06-28 13:13:40 +0100]:
> Mark M. Hoffman wrote:
> > I skimmed through the blog entry and the earlier one which it references.
> > In your first attempt you reinvented the 'P' part of a PID control loop.
> > Don't give up! ;)  There's a *ton* of info about PID controllers on the
> > web.  If you're still interested, you could start here:
> > 
> > http://en.wikipedia.org/wiki/PID_controller
> 
> Very interesting stuff, thanks!
> 
> My quick reading of this stuff indicates a possible problem.  On my 
> system at least, I can only read the temperature in whole degrees; 
> lm-sensors returns a number in thousandths of a degree, but sadly the 
> last three digits are always zero.  This means that our differentiation 
> of the temperature will be very crude, which could make it difficult to 
> build a functioning PID controller.

Read under 'Limitations' in the wikipedia article.  One suggestion is
to just skip the differential band altogether (PI controller).

> How can I find out the maximum safe operating temperature of my current 
> CPU?  Since my first priority is something that works on my current 
> system, the only reason for me to pick this up again and explore in more 
> detail is if I can afford to slow the fan even when the system is flat 
> out, which means I need to know how fast I can run it.

You can find this info in processor datasheets, which are available on
Intel/AMD websites.  It shouldn't be too hard to find.  Otherwise, you
could consider 55 C to be a very conservative limit.  Some Intel cpus
are OK up to 85 C, but it really does depend on the model *and* maybe
even the stepping.

Something else to consider: if you do tune your system to keep the CPU
near max temp, make sure that the extra heat is not going to affect
your disk(s).

Regards,

-- 
Mark M. Hoffman
mhoffman at lightlink.com

[lm-sensors] A better fan control program

[Jean Delvare]

Hi Paul,

> How can I find out the maximum safe operating temperature of my current 
> CPU?  Since my first priority is something that works on my current 
> system, the only reason for me to pick this up again and explore in more 
> detail is if I can afford to slow the fan even when the system is flat 
> out, which means I need to know how fast I can run it.

There's this this page I use as a reference for CPU characteristics:
http://users.erols.com/chare/elec.htm

You can also find the information the datasheet of your CPU. Both Intel
and AMD provide this information, if you search long enough.

Keep in mind that the temperature reported by lm_sensors (or any other
mean) is almost always an approximation, either because of the
temperature probe (lack of) accuracy, or because the probe is not
inside the CPU but just nearby, or because the the measurement noise,
or because the driver caches the data for a certain amount of time,
etc. So you should keep a good safety margin. Colder is always better
for the lifetime of your CPU, as you found out yourself.

-- 
Jean Delvare