Page 1 of 1
core 17 temperature control with nvidia under linux
Posted: Mon Aug 31, 2015 10:00 pm
by mattifolder
I have configured my folding client under linux as described there
https://folding.stanford.edu/home/chang ... -full-fah/ or there
viewtopic.php?f=74&t=26533&p=266854#p266854. The Log shows the following configuration:
Code: Select all
...
19:32:19: OS: Linux 3.19.0-26-generic x86_64
19:32:19: OS Arch: AMD64
19:32:19: GPUs: 1
19:32:19: GPU 0: NVIDIA:5 GM204 [GeForce GTX 970]
19:32:19: CUDA: 5.2
19:32:19:CUDA Driver: 7050
...
19:48:35:<config>
...
19:48:35: <!-- Folding Slot Configuration -->
19:48:35: <extra-core-args v='-forceasm -tmax=70 -twait=900'/>
...
19:48:35: <!-- Folding Slots -->
...
19:32:19: <slot id='1' type='GPU'>
...
19:32:19: <extra-core-args v='-tmax=70 -twait=900'/>
...
19:32:19: </slot>
19:48:35:</config>
Now with Core 17, 18 or 21 projects it seems that the options are not accepted. The process starts with theses options and is listed so in process list, but there is also a warning logged, that temperature control is disabled:
Code: Select all
19:34:08:WU00:FS01:Connecting to 171.67.108.200:80
19:34:08:WU00:FS01:Assigned to work server 171.64.65.98
19:34:08:WU00:FS01:Requesting new work unit for slot 01: READY gpu:0:GM204 [GeForce GTX 970] from 171.64.65.98
19:34:08:WU00:FS01:Connecting to 171.64.65.98:8080
19:34:09:WU00:FS01:Downloading 7.48MiB
19:34:12:WU00:FS01:Download complete
19:34:12:WU00:FS01:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:9712 run:67 clone:32 gen:18 core:0x21 unit:0x00000067ab40416255b9b07824c8d8d6
19:34:12:WU00:FS01:Starting
19:34:12:WU00:FS01:Running FahCore: /usr/bin/FAHCoreWrapper /var/lib/fahclient/cores/web.stanford.edu/~pande/Linux/AMD64/NVIDIA/Fermi/Core_21.fah/FahCore_21 -dir 00 -suffix 01 -version 704 -lifeline 2243 -checkpoint 3 -gpu 0 -gpu-vendor nvidia -tmax=70 -twait=900
19:34:12:WU00:FS01:Started FahCore on PID 3346
19:34:12:WU00:FS01:Core PID:3350
19:34:12:WU00:FS01:FahCore 0x21 started
19:34:12:WU00:FS01:0x21:*********************** Log Started 2015-08-31T19:34:12Z ***********************
19:34:12:WU00:FS01:0x21:Project: 9712 (Run 67, Clone 32, Gen 18)
19:34:12:WU00:FS01:0x21:Unit: 0x00000067ab40416255b9b07824c8d8d6
19:34:12:WU00:FS01:0x21:CPU: 0x00000000000000000000000000000000
19:34:12:WU00:FS01:0x21:Machine: 1
19:34:12:WU00:FS01:0x21:Reading tar file core.xml
19:34:12:WU00:FS01:0x21:Reading tar file integrator.xml
19:34:12:WU00:FS01:0x21:Reading tar file system.xml
19:34:13:WU00:FS01:0x21:Reading tar file state.xml
19:34:13:WU00:FS01:0x21:Digital signatures verified
19:34:13:WU00:FS01:0x21:Folding@home GPU Core21 Folding@home Core
19:34:13:WU00:FS01:0x21:Version 0.0.11
19:34:35:WU00:FS01:0x21:Completed 0 out of 1280000 steps (0%)
19:34:35:WU00:FS01:0x21:Temperature control disabled. Requirements: single Nvidia GPU, tmax must be < 110 and twait >= 900
...
Is this function not activated under linux or is my configuration not correct ?
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 2:22 am
by 7im
This is a very crude tool of last resort. You would be better off trying to use cool bits to manage temps to a finer grain.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 2:59 am
by bruce
davidcoton wrote:
sudo nvidia-xconfig --cool-bits=12
reboot
Then fan control and clocking adjustments are available in nvidia-settings
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 6:15 am
by mattifolder
I know cool-bits, nvidia-settings and nvidia-smi and already use it to adjust gpu clocks. Btw. there are temporarely high room temperatures, that not may to be planned. Scripted automation would produce an extra cpu overhead and riscs of instability. So I thougt, tmax and twait may be a simple solution. I only want to know, if it has to be functionally or not.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 10:19 am
by ChristianVirtual
To my understanding chips don't like frequent jump in temperature. Rather constant temps is what they can deal better with.
And even if you have a waiting time of whatever seconds, as soon the folding continues the temp rise again.
Better to ensure sufficient ambient temps/airflow; or if you want to get fancy replace the fans with a water-block
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 2:41 pm
by mattifolder
May someone please answer my question: are extra-core-args tmax and twait functionally under linux ?
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 4:40 pm
by 7im
From the core 17 thread.
Code: Select all
This is a feature present in FahCore_17 version .52 and higher.
Until they release a newer version of the present Linux core (.46) which is unlikely since they have already moved on to core 18 and 21, the feature is not supported in Linux. As I posted earlier, coolbits is a better option, assuming a workable environment temp.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 5:28 pm
by mattifolder
Ok, thank you. That's just now an clearly answer. I thought, I had also core 17 projects under linux. But in my saved logs only core 18 (version 0.0.4) and 21 (version 0.0.3). One critically opinion I've so long, such log entries are bewildering:
Code: Select all
19:34:35:WU00:FS01:0x21:Temperature control disabled. Requirements: single Nvidia GPU, tmax must be < 110 and twait >= 900
In my system that actually means:
Temperature control disabled, because (although) I have an Nvidia GPU and configured tmax < 119 and twait >= 900.
I'm developing software by myself at business and I hate
programs that seem doing things, that they don't do.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 6:53 pm
by 7im
I do not know if this feature works in core 18 or 21, or in what versions of those cores.
But turning off folding for 5 minutes (900 seconds) or more was wasteful, especially when checkpoints are saved every 5 frames (%). Each 5 minute shutdown could cost several minutes to several hours of folding, depending on average frame times, and when the last checkpoint was saved to disk.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 8:18 pm
by davidcoton
mattifolder wrote: I hate
programs that seem doing things, that they don't do.
So do most of us. As I found when I was involved in software development, getting it right is not always the management priority
In this case software development is overloaded with stuff to benefit the science plus a lot of recent work on the server infrastructure. There are lots of bits left over from the initial v7 development (up to 7.4.4) which are not quite right, but have not been assigned enough priority to get them fixed -- there is no announced intention to release a 7.5.x client. That misleading message is probably not even high amongst the known issues
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 8:37 pm
by mattifolder
7im wrote:But turning off folding for 5 minutes (900 seconds) or more was wasteful
Oh man, twait are seconds !
Who reads correctly has really an advantage.
I really thought, that are milliseconds. You can forget (close) the thread. I'll fast deactivate that configuration option and solve my problem with appropriate reduced OC and cooling of the system when the summer is too hot. At the moment it nearly reaches to reduce GPU Core Clock of 5 to 10 Mhz and / or suspend CPU Folding.
Re: core 17 temperature control with nvidia under linux
Posted: Tue Sep 01, 2015 10:29 pm
by 7im
Ja, Feinabstimmung und nicht die Grobabstimmung.