Automatic Temperature regulation

log in

Advanced search

Message boards : Questions and problems : Automatic Temperature regulation

1 · 2 · 3 · 4 . . . 11 · Next
Author Message
Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 22392 - Posted: 12 Jan 2009, 14:45:30 UTC

I made a program that monitors the Processor core temperature and throttles a running task so that the temperature stays within limits. It works on Windows 32 Bit Intel only. But if there are AMD users around who want to help… Processor needs to have a build in sensor and cpu register, so most computers up to 2 years ago will work. I made this specifically for my laptop.Download
This is a beta program, but tested on several computers (XP/VISTA). Suggestions welcome.

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 22485 - Posted: 16 Jan 2009, 17:35:42 UTC - in response to Message 22392.

AMD users: I'm looking for testers. Even if this message is old please contact me.
A new version 1.1 with automatic program detection is ready for download.

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 22717 - Posted: 27 Jan 2009, 10:50:02 UTC - in response to Message 22485.

A new version 1.4 is available.
You can make a graph of the temperature.For 1 or all processor cores.
This allows monitoring from minutes to the whole day.
Download this free program

Pepo
Avatar
Send message
Joined: 3 Apr 06
Posts: 547
Message 22897 - Posted: 5 Feb 2009, 9:47:56 UTC

Although I've responded to Fred per PM a few days ago, I've decided to write my comments also here, to stimulate development :-)


  • V1.4 immediately crashes on 600 MHz AMD Duron. I admit my CPU lacks "a build in sensor and cpu register", but still, it should not behave that harsh to the old uncle ;-) (Fred is already looking into detecting and using AMD regs correctly.)
  • On my C2D T7200 it works as advertised. I appreciate the large temperature graphs, I was looking for ages for something like this!
  • I think that especially multiproject BOINC crunchers would appreciate a possibility to enter not just single executables (which get occasionally replaced and renamed), but whole path roots, like ...\BOINC\projects\ or ...\BOINC\slots\, otherwise each "new" application will not get throttled.
  • Beware! At least the TThrottle 1.4 incorrectly shows its name as "TTrottle", which is nearly similar to a german swear-word :-)



Peter

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 22898 - Posted: 5 Feb 2009, 10:47:21 UTC - in response to Message 22897.

Although I've responded to Fred per PM a few days ago, I've decided to write my comments also here, to stimulate development :-)

As we speak we are now testing 1.43 with AMD users. (and making good progress)
Seti AMD users
That with the automatically detect on BOINC task is a good idea. Will surely add it.
Making the name shorter helps sometimes putting in Einstein will throttle any Einstein task stating with this name.

Pi3
Send message
Joined: 14 Jan 09
Posts: 6
Message 22917 - Posted: 6 Feb 2009, 8:49:19 UTC

Hi, very cool program - if you get the joke - !!

Really it looks very professional and hopefully it will help with my automatic shut off problems.

I was wandering if something could be done to make it easier (i.e. foolproof and let's say automatic) to detect the processes it monitors and regulates: many of the applications running on boinc are frequently updated and this makes it necessary to go and update the list. Plus, many boinc users like to run a variety of projects, so this makes it a long task to load them all in TT.

This is just a couple of thoughts to beef up the wishlist, I am sure you are already working on it.

Speaking of wishlist, it may almost be a good idea to implement your program IN BOINC, or otherwise work on connecting the two programs some more.

Keep up the good work!!!

Pepo
Avatar
Send message
Joined: 3 Apr 06
Posts: 547
Message 22921 - Posted: 6 Feb 2009, 12:15:00 UTC - in response to Message 22917.

I was wandering if something could be done to make it easier (i.e. foolproof and let's say automatic) to detect the processes it monitors and regulates: many of the applications running on boinc are frequently updated and this makes it necessary to go and update the list. Plus, many boinc users like to run a variety of projects, so this makes it a long task to load them all in TT.

This would be exactly covered by automatically regulating each and any application, running under the BOINCprojects path tree. Maybe in this case it would be good to implement exceptions from such automatism?

Speaking of wishlist, it may almost be a good idea to implement your program IN BOINC, or otherwise work on connecting the two programs some more.

Exactly

Peter

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 22967 - Posted: 9 Feb 2009, 10:25:52 UTC - in response to Message 22921.

I was wandering if something could be done to make it easier (i.e. foolproof and let's say automatic) to detect the processes it monitors and regulates: many of the applications running on boinc are frequently updated and this makes it necessary to go and update the list. Plus, many boinc users like to run a variety of projects, so this makes it a long task to load them all in TT.

This would be exactly covered by automatically regulating each and any application, running under the BOINCprojects path tree. Maybe in this case it would be good to implement exceptions from such automatism?

Speaking of wishlist, it may almost be a good idea to implement your program IN BOINC, or otherwise work on connecting the two programs some more.

Exactly

Peter

New version that does all that. Download 1.45

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 23128 - Posted: 15 Feb 2009, 15:03:31 UTC - in response to Message 22967.

A new version that includes 64Bit support (tested on XP64 / 2003) download v 1.47

Pi3
Send message
Joined: 14 Jan 09
Posts: 6
Message 23139 - Posted: 16 Feb 2009, 14:24:04 UTC

Hi,

you guys are doing a great job, the program seems to be making good progress!

One thing I was wondering now, is:

Is there some sort of rule of thumb to mantain a reasonably efficient run time in terms of FLOPS per Kilowatt hour?

Any way to set let's say a maximum core temperature in order to maximize computation with the least power consumption? Or are they simply directly proportional (so if i double run time, computation ouput will grow twice as much and so will be consumption)

I am not sure I am making myself clear, or that I am making any sense at all!

Please tell me if this has been covered elsewhere. In this times of greater concern for the environment, I feel BOINC should take into account that an effort to reduce consumption may look attractive to potential crunchers that have ecologic (and -nomic) worries.

Thanks

Sekerob
Send message
Joined: 25 Aug 06
Posts: 1398
Message 23158 - Posted: 17 Feb 2009, 8:20:59 UTC - in response to Message 23139.
Last modified: 17 Feb 2009, 8:38:55 UTC

Been running it for past 48 hours on a quad. It seems the throttling is across the board rather than core specific...am I right? At least when running a combo of WCG Rice and D@H and watching in Process Explorer it's torquing off all 4 processes. I know D@H is the one raising temps about 4C as does Primergrid and QMC, so controlling at core level would seem a way to keep other sciences unperturbed.

For the graph, would be nice to see which science is using what core.

Thanks for the excellent addition of tools to optimize crunching.

PS: I'm using ThreadMaster GUI to set a percent per science during summer, but temperature controlled use at science level would I think be the better solution for many and use maximum CPU still as the ambient temperature fluctuates.
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 23165 - Posted: 17 Feb 2009, 11:33:38 UTC - in response to Message 23158.

Been running it for past 48 hours on a quad. It seems the throttling is across the board rather than core specific...am I right? At least when running a combo of WCG Rice and D@H and watching in Process Explorer it's torquing off all 4 processes. I know D@H is the one raising temps about 4C as does Primergrid and QMC, so controlling at core level would seem a way to keep other sciences unperturbed.

For the graph, would be nice to see which science is using what core.

Thanks for the excellent addition of tools to optimize crunching.

PS: I'm using ThreadMaster GUI to set a percent per science during summer, but temperature controlled use at science level would I think be the better solution for many and use maximum CPU still as the ambient temperature fluctuates.

Indeed TThrottle does affect all project on any core.
The main problem is that the OS distributes work over the cores. And not only at the beginning and end of a task. throttling all projects should give a more or less equal distribution of work. But as not all work is equal... Some work may increase the temp more then others. And the reading may not be too precise. Intel e.g. gives a 1C tolerance and the converter may be 1C off as well. That could mean if you see a 4C difference in core temp it may be still be within normal tolerance. Meaning what you see is not what happens and the cores may have exactly the same temperature. If there are enough request I may add a feature to limit throttling to specific project only, if that is what you mean.

Sekerob
Send message
Joined: 25 Aug 06
Posts: 1398
Message 23166 - Posted: 17 Feb 2009, 12:06:35 UTC - in response to Message 23165.
Last modified: 17 Feb 2009, 12:36:31 UTC

There's already a science exclude option, but then the throttling would only work if the specific "hot" were running when the objective is to maintain a maximum applying to all cores whilst even in summer the RICE would cause the CPU to get too hot at 100% load unless setting Speedfan to cause the curtains to wave.

As to tolerance, been playing with Process Lasso to quickly move affinities and temps very closely follow the core where D@H is running on. Apps as CoreTemp and Speedfan show same, so I'm reasonably confident that the temps are well represented, much closer than Intel is saying.

Anyway, just a thought to make the control science specific. On a Quad easily 5-10% of computing power is lost because of a very hot running science. ThreadMaster still allows to reign in the hottest individually, but TThrottle would be the ideal all-in-one control, set and forget 365 day of the year.

cheers

Added: Other thought, considering the tolerance, running the control on a composite temp of all cores might be a way instead of the absolute measured for any core, but that would only be a minimal gain I think.
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt

benDan
Avatar
Send message
Joined: 9 Feb 09
Posts: 22
Message 23174 - Posted: 17 Feb 2009, 22:20:09 UTC
Last modified: 17 Feb 2009, 22:32:54 UTC

I am using the beta (1.4.7) version.
I now have a functioning throttel.
Boinc info is now shown.
I can now run BOINC tasks unattended.
THANK GOD!

I tried both ReadyDriverPlus
and Driver Signature Enforcement Overrider 1.2
from http://www.ngohq.com/home.php?page=dseo

Both work but I settled with Overrider.


system is:
ASUS M2N-SLI Deluxe
AMD Athlon(tm) 64 X2 Dual Core Processor 6000+
Microsoft® Windows Vista Home Premium 64

benDan
Avatar
Send message
Joined: 9 Feb 09
Posts: 22
Message 23193 - Posted: 18 Feb 2009, 16:44:08 UTC

Changed this AM to the latest beta (1.4.8) version.

I stopped using Driver Signature Enforcement Overrider 1.2 yesterday because of failure to stay in test status.
I went back to ReadyDriverPlus and it is still working.
I see no real difference except it now shows milkyway.cs.rpi.edu_milkyway in the BOINC list.

Fred: ... do want a current log?

Thanks again
____________
--
benDan

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 23195 - Posted: 18 Feb 2009, 17:13:26 UTC - in response to Message 23193.

Changed this AM to the latest beta (1.4.8) version.
I see no real difference except it now shows milkyway.cs.rpi.edu_milkyway in the BOINC list.

The changes involve running the 32 bit version on a 64 bit system and redirecting the user with a warning. Made the 1e program tab simpler by removing most of the unused buttons etc and placed an expert check for those who want it anyway. Makes some room for the list. Most early users can now (normally) remove all programs from the 1e program tab, by selecting and "remove". Fixed the BOINC detection in the X64 version. And added additional AMD processor info.

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 23266 - Posted: 24 Feb 2009, 12:58:57 UTC - in response to Message 23195.

New version 1.51: Allows monitoring GPU (CUDA) temperature. A separate throttle can keep the GPU temperature at the set temperature.
The graph shows the temperature drop as one task ends and another task is starting. Another interesting thing came up. The cpu tasks that communicates with the GPU needs to run at about 70% of time to fully fill up the GPU with work.

Sekerob
Send message
Joined: 25 Aug 06
Posts: 1398
Message 23269 - Posted: 24 Feb 2009, 13:47:46 UTC - in response to Message 23266.

Fred,

Running 1.47 for longer have had several times a "Quota 100 reached ..." type of system instability. Launching, saving data, became impossible. Killing TThrottle restored system stability.
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt

Fred - efmer.eu/boinc
Avatar
Send message
Joined: 8 Aug 08
Posts: 529
Message 23270 - Posted: 24 Feb 2009, 13:57:46 UTC - in response to Message 23269.

Fred,Running 1.47 for longer have had several times a "Quota 100 reached ..." type of system instability. Launching, saving data, became impossible. Killing TThrottle restored system stability.

Try 1.51, I made some memory changes as well.

Sekerob
Send message
Joined: 25 Aug 06
Posts: 1398
Message 23283 - Posted: 24 Feb 2009, 19:10:32 UTC - in response to Message 23270.

Hi Fred,

Gave 1.51 a spin and after 4 hours, the Firefox app with 3 longer writings, not yet saved, disappeared off screen. Trying to relaunch gave the same relaunch error... Quota all used, free resources... Again when ending TThrottle, the temp icon (coretemp), went from 100, which I think is the junction point, to the regular op temp.

Thanks
____________
Coelum Non Animum Mutant, Qui Trans Mare Currunt

1 · 2 · 3 · 4 . . . 11 · Next
Post to thread

Message boards : Questions and problems : Automatic Temperature regulation


BOINC home page · Log in · Create account

Copyright © 2014 University of California. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.