Advanced search

Message boards : Number crunching : GPUGRID and Linux

Author Message
TJ
Send message
Joined: 26 Jun 09
Posts: 815
Credit: 1,470,385,294
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33586 - Posted: 23 Oct 2013 | 8:36:06 UTC

Hello,

I am planning to do a bit of GPUGRID with Linux in the future (as I know how ti install BOINC in the way that I have control over it and have access to the directories and files (which I have not when let Linux instal BOINC automatically)), but I see that the stderr file with all information MJH made is only available in Windows.
Are there no programs in Linux to measure system temperature and such?
____________
Greetings from TJ

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33592 - Posted: 23 Oct 2013 | 14:35:11 UTC - in response to Message 33586.
Last modified: 23 Oct 2013 | 14:42:15 UTC

Generally there is a lack of such sensor readers on Linux, however there is just about enough to get by on.

FAQ - Useful Tools

On Ubuntu, you can use System Monitor to see the CPU usage (core by core) - It comes with the OS.
You can install Psensor to see the temperature of the CPU and GPU's.
NVidia X Server (installed with the NVidia drivers) tells you the GPU temperature, how fast the GPU fan is, and allows you to change the fan speed and set to prefer maximum performance - if you can convince it to work using coolbits.

Matt indicated in the forum that he would do some work on the Linux app to enhance the stderr output in the future.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33593 - Posted: 23 Oct 2013 | 17:24:38 UTC - in response to Message 33592.

Hello: One of the best ways to control the system possibly be "Gkrellm" capable of displaying all values​​; processor usage, Temperatures (CPU and GPU) fans, Processes etc. ..

I have it installed on Ubuntu but it sure works with any distribution. Greetings.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33635 - Posted: 27 Oct 2013 | 11:35:23 UTC - in response to Message 33593.

How do you get it to show GPU temperature or fan speed?
Can it be used to control GPU fan speed?
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33666 - Posted: 29 Oct 2013 | 19:17:11 UTC - in response to Message 33635.

Hello: control fan speed on the GPUs is through "Coolbits" in Nvidia-Settings, the new version shows the fan speed of the GPU if it is supported.

Gkrellm not show fan speed of the GPUs but if their temperatures naturally first run "lm-sensors" to detect all sensors and well plate Gkrellm (or any other monitoring utility) can read them.

If anyone is interested I can detail the steps to activate "Coolbits" run "lm-sensors" and mount "Gkrellm". all this for Ubuntu 13.10 but I guess it is for other distributions. Greetings.

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33669 - Posted: 29 Oct 2013 | 19:44:54 UTC - in response to Message 33666.
Last modified: 29 Oct 2013 | 20:00:28 UTC

Carlesa, that would be useful.

I created a new thread called coolbits, within the Number crunching area.

Kindly post your How To use coolbit with Ubuntu 13.10 there.

Others can add their methods for using coolbits with different versions of Linux, under different conditions (like a headless setup), tips and experiences using coolbits with single or multiple cards...

COOLBITS

Thanks,
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33676 - Posted: 30 Oct 2013 | 12:51:50 UTC

I've seen reports at Einstein that the latest 331 linux beta drivers finally report some more diagnostic information, even listing GPU usage.

MrS
____________
Scanning for our furry friends since Jan 2002

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33678 - Posted: 30 Oct 2013 | 13:23:57 UTC - in response to Message 33676.

Hello: I'm using Nvidia 331.17 for days and the basic difference with the above is that it shows the fan speed of the GPU and this supports it.

No information about the load or use in% of the GPUs, if the frequency of the graphics and memory, I can post images of Nvidia-Settings if interested.

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33679 - Posted: 30 Oct 2013 | 14:57:44 UTC - in response to Message 33678.
Last modified: 30 Oct 2013 | 15:15:15 UTC

Hello: I'm using Nvidia 331.17 for days and the basic difference with the above is that it shows the fan speed of the GPU and this supports it.

No information about the load or use in% of the GPUs, if the frequency of the graphics and memory, I can post images of Nvidia-Settings if interested.



Hello: Sorry I had misunderstood.

If that information is the use of GPUs.

% Use of the GPU
% Use of PCIe Bandwidth
% Use video engine

At the moment I can not represent you in an outer panel GKrellm type. Greetings.

Nvidia-Settings- Use % GPU... etc.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33732 - Posted: 3 Nov 2013 | 4:42:31 UTC - in response to Message 33586.

Hello,

I am planning to do a bit of GPUGRID with Linux in the future (as I know how ti install BOINC in the way that I have control over it and have access to the directories and files (which I have not when let Linux instal BOINC automatically)), but I see that the stderr file with all information MJH made is only available in Windows.
Are there no programs in Linux to measure system temperature and such?


I run GPUgrid exclusively on Linux so I'm afraid I have no idea what information MJH makes available in stderr.

Linux has apps for measuring system temp and the Linux nvidia drivers provide, IMHO, all the info one could possibly want about nvidia GPUs running in the system. The nvidia-settings binary is the interface to driver. It reads and reports state (temp, %usage, freq, etc.) and allows altering parameters such as fan speed, freq, etc. The problem with nvidia-settings is that it is pretty much a CLI application. If you run it with no command line parameters it pops up a GUI interface but that interface is rather limited, IMHO. If you run it with command line parameters it does not invoke the GUI interface and stays in CLI mode. You could say it has a split-personality disorder. Also, the CLI mode is extremely cryptic.

I'm not sure how everybody else feels about it but IMHO the biggest problems with the nvidia drivers on Linux and the nvidia-settings interface are:

1. You can use the interface to either overclock or control fan speed but not both. If you setup to overclock (with Coolbits 1) then nvidia takes away the fan speed part of the interface and runs the fan speed on auto. You can read the fan speed but not adjust it. If setup for manual fan speed adjustment (with Coolbits 4) then you can read freqs and volts but you cannot adjust them.

2. When the fan speed is in auto mode (i.e. you have used Coolbits 1 or have not set Coolbits at all) the user has no control over the threshold temperature (aka target temp). In auto mode the driver keeps the temp hovering between 82C to 85C and sometimes it spikes to 88C on my system. I would like it to run cooler, somewhere closer to 70C.

To overcome 1, you can hack the card and remove the PWM control line to make the fan run at 100% all the time. Then you can set Coolbits 1, overclock, and hopefully not go above 80C temperature.

Another way to overcome 1 is to build a programmable PWM fan speed controller and use it to control fan speed the way YOU want it controlled. I've been checking that out and it's doable for about $20 (parts only, no labor).

The fixed threshold temperature I mention in 2 above can be overcome by setting Coolbits 4 and setting the fan to a high speed. That works if the extra noise is not a problem. If the noise is a problem then it is possible to write a front end to the nvidia-settings app that allows you to specify whatever threshold temp you wish and auto adjusts fan speed to maintain that threshold. I wrote such a front end in Python, it worked very well, but later I simply detached the fan from the onboard speed controller and let it run at 100%.

I don't know what MJH has planned. Tell me what kind of info you would like to see and I might program it for you. Or, if you can code and want to roll your own then I might be able to get you pointed in the right direction re: tools, docs, etc.

____________
BOINC <<--- credit whores, pedants, alien hunters

Profile [VENETO] sabayonino
Send message
Joined: 4 Apr 10
Posts: 50
Credit: 645,641,596
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33733 - Posted: 3 Nov 2013 | 7:25:13 UTC

Hi

latest nvidia-settings (331.17) and drivers provide more informations about your graphic card

nvidia-settings

see

$ nvidia-settings --help


nvidia-smi (CLI-Mode)

see
$ nvidia-smi --help


regards

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33740 - Posted: 3 Nov 2013 | 12:23:39 UTC - in response to Message 33733.

Hello Dagorath, in these times the last driver 331.17 Linux (Beta) and Nvidia-Settings set to "Coolbits 4" you have access to the necessary information from the GPU and a very good manual fan control with nothing to envy options in Windows.

Automatic programming missing fan according to the temperature forcing some temperatures monitor the GPUs we can get screen continuously and scheduled alarms, for example com GKrellm, as I said at another time.

The issue I Coolbits widely discussed in this section in another thread. Greetings.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33749 - Posted: 3 Nov 2013 | 16:16:15 UTC - in response to Message 33741.
Last modified: 3 Nov 2013 | 16:16:39 UTC

Hi Carlesa,

I installed the 331.17 (beta) driver and yes there is more information in the nvidia-settings GUI. There is also the new rules and profile section. I have Coolbits 4 working and I have the manual fan speed control working.

Maybe I missed something but the manual fan speed control has not improved since earlier drivers and the auto fan speed control still does not allow setting a lower target temperature. Maybe I'm doing it wrong.

Thanks for your suggestion about GKrellM. I didn't know about it until you mentioned it. I like it and the alarms are nice but it seems to me the alarms only issue a warning and that is all. If the alarms could also trigger an action such as a fan speed increase they would be more useful.
____________
BOINC <<--- credit whores, pedants, alien hunters

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33751 - Posted: 3 Nov 2013 | 17:43:49 UTC - in response to Message 33749.

Hello: The manual fan control by Coolbits 4 works perfectly, you can look at the thread "Coolbits" to review how to install.

The problem is that it only allows you to set the fan manually % but is not able to control automatically change according to temperature, you have to monitor possible changes in temperature and manually reset.

It works whether you have one or more GPUs installed, but see problem connected monitor.

GKrell have visual alarms but can not control the fans that I know.

For GKrell (actually any monitor utility) works well and read all the sensors on your motherboard first install and run "lm-sensors". Greetings.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33753 - Posted: 3 Nov 2013 | 18:36:48 UTC - in response to Message 33751.
Last modified: 3 Nov 2013 | 18:59:08 UTC

Hello: The manual fan control by Coolbits 4 works perfectly, you can look at the thread "Coolbits" to review how to install.

The problem is that it only allows you to set the fan manually % but is not able to control automatically change according to temperature, you have to monitor possible changes in temperature and manually reset.


You can manually reset or you can run a script that specifies a target temperature (eg. 70C) and calls

nvidia-settings --query localhost:0[thermalsensor:0]/ThermalSensorReading


to query the current temp. If current temp > target temp then call

nvidia-settings --assign localhost:0[fan:0]/GPUCurrentFanSpeed=X


where X = an interger > current fan speed

If current temp < target temp then call

nvidia-settings --assign localhost:0[fan:0]/GPUCurrentFanSpeed=X


where X = an interger < current fan speed.

You can programatically switch from manual fan control to auto fan control with

nvidia-settings --assign localhost:0[gpu:0]/GPUFanControlState=1


Pass 0 instead of 1 to the above command to switch back to auto fan control.

Run the commands in a loop for continuous auto temp control to your target temp not nvidia's target temp.

edit: spelling correction, localhst to localhost
____________
BOINC <<--- credit whores, pedants, alien hunters

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33755 - Posted: 3 Nov 2013 | 19:00:42 UTC - in response to Message 33753.

Hello Dagorath thanks for the interesting information new to me, I will try to use it if I can.

My GTX770 for now works great and fresh as I have configured with a load of 93% <60 ° C and <65% fan speed, but it is interesting to try to automate the process, we'll see. Greetings.

Dagorath
Send message
Joined: 16 Mar 11
Posts: 509
Credit: 179,005,236
RAC: 0
Level
Ile
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33756 - Posted: 3 Nov 2013 | 20:21:13 UTC - in response to Message 33755.

Hi Carlesa,

I already have it automated with a bash script. Unfortunately I lost that script but I'll rewrite it as a Python script and make it available.

The nice thing about automating it with a script is that you can specify the target temp plus a maximum fan speed. If the current temp exceeds the target temp and current fan speed = max fan speed then the script can reduce % usage or clock speed to reduce the running temp.


____________
BOINC <<--- credit whores, pedants, alien hunters

Profile Carlesa25
Avatar
Send message
Joined: 13 Nov 10
Posts: 328
Credit: 72,619,453
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 33758 - Posted: 3 Nov 2013 | 20:53:45 UTC - in response to Message 33756.

Hi Carlesa,

I already have it automated with a bash script. Unfortunately I lost that script but I'll rewrite it as a Python script and make it available.

The nice thing about automating it with a script is that you can specify the target temp plus a maximum fan speed. If the current temp exceeds the target temp and current fan speed = max fan speed then the script can reduce % usage or clock speed to reduce the running temp.



Hello: It would be great facilitates things, be good to use their experience, I will watch thanks. Greetings.

Post to thread

Message boards : Number crunching : GPUGRID and Linux

//