NVidia-Linux Adjustments for heat

Message boards : Number crunching : NVidia-Linux Adjustments for heat
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Profile Retvari Zoltan
Avatar

Send message
Joined: 20 Jan 09
Posts: 2380
Credit: 16,897,957,044
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52352 - Posted: 26 Jul 2019, 21:25:01 UTC - in response to Message 52351.  

I wish I knew how to write a script that would shutdown at 1:00 and power on at 6:00.
Shutdown: https://forums.linuxmint.com/viewtopic.php?t=22251
Power on: it can be configured in the BIOS. You should check your local time, and the BIOS clock, as Linux tends to set GMT (=UTC) in the BIOS, so the wake up time should be set according to this.
ID: 52352 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile JStateson
Avatar

Send message
Joined: 31 Oct 08
Posts: 186
Credit: 3,578,903,157
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52353 - Posted: 27 Jul 2019, 3:00:26 UTC

re: my problem with only 5 out of 7 GPUs having fan control.

This was unexpectedly fixed when I swapped a pair of video boards. There must be some type of timing problem with nvidia-settings when a lot of GPUs are on a motherboard. I have an assortment, cheap from eBay & used, different vendors, pair of 1070 and five 1060 one of which is 6gb the others 3 and simply swapping the USB riser cable on the two 1070 boards was enough to enable the temperature slider on all 7 boards. One of the sliders I had to set to 100 the other 6 were already at 100. The 2 I swapped were different models. Should have made no difference but yet it did.
ID: 52353 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Keith Myers
Avatar

Send message
Joined: 13 Dec 17
Posts: 1419
Credit: 9,119,446,190
RAC: 731
Level
Tyr
Scientific publications
watwatwatwatwat
Message 52354 - Posted: 27 Jul 2019, 7:31:49 UTC

The posts from the high-gpu count Seti users that have multiple gpus in mining type motherboards usually comes down to fixing the problems with replacing the USB cables from the risers with higher quality shielded cables.

All sorts of problems ranging from gpus disappearing to gpus with limited control seem to be fixed with replacement of the USB cables.
ID: 52354 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
biodoc

Send message
Joined: 26 Aug 08
Posts: 183
Credit: 10,085,929,375
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52355 - Posted: 27 Jul 2019, 9:57:15 UTC

Has anyone tried using nfancurve?

"A small and lightweight POSIX script for using a custom fan curve in Linux for those with an Nvidia GPU."

https://github.com/nan0s7/nfancurve

Another set of instructions for use.

https://www.techticity.com/howto/how-to-control-nvidia-graphics-card-fan-speed-in-linux/
ID: 52355 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
rod4x4

Send message
Joined: 4 Aug 14
Posts: 266
Credit: 2,219,935,054
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 52356 - Posted: 27 Jul 2019, 16:59:34 UTC - in response to Message 52355.  

looks promising. The "to do" list on the first link does mention support still to be added for headless applications. So limited in that respect. Still looking for a fully scripted solution for a headless environment.
ID: 52356 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile tito

Send message
Joined: 21 May 09
Posts: 22
Credit: 2,002,780,169
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwat
Message 52357 - Posted: 27 Jul 2019, 19:26:57 UTC - in response to Message 52356.  

I have checked that scirpt - it works good for 1 GPU 1080Ti (sorry don't have more to check).
ID: 52357 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Keith Myers
Avatar

Send message
Joined: 13 Dec 17
Posts: 1419
Credit: 9,119,446,190
RAC: 731
Level
Tyr
Scientific publications
watwatwatwatwat
Message 52358 - Posted: 28 Jul 2019, 3:12:23 UTC

I'd like to figure out how nvidia-settings uses the fan control interface identifier on the Turing cards to differentiate the two interfaces.

I have a fan control application with a nice GUI interface written in Python, that does not work on the Turing cards with the two interfaces. The first interface is enabled and can be controlled, but the second interface is missing. I've looked through the code and can't figure out what the variable name is for the fan interface. I thought it would be simple enough to increment the variable for each interface found, but nothing is obvious to me.

The best part of the application is that is has curves and not just static levels like the fan speed slider in Nvidia X Server Settings app.

ID: 52358 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
rod4x4

Send message
Joined: 4 Aug 14
Posts: 266
Credit: 2,219,935,054
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 52359 - Posted: 28 Jul 2019, 6:58:56 UTC - in response to Message 52358.  

I'd like to figure out how nvidia-settings uses the fan control interface identifier on the Turing cards to differentiate the two interfaces.


Does the below link help? I dont have a Turing based card so cant test this.
https://linustechtips.com/main/topic/1048251-turing-rtx-linux-cli-fan-control/
ID: 52359 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
biodoc

Send message
Joined: 26 Aug 08
Posts: 183
Credit: 10,085,929,375
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52360 - Posted: 28 Jul 2019, 10:26:29 UTC - in response to Message 52359.  

I'd like to figure out how nvidia-settings uses the fan control interface identifier on the Turing cards to differentiate the two interfaces.


Does the below link help? I dont have a Turing based card so cant test this.
https://linustechtips.com/main/topic/1048251-turing-rtx-linux-cli-fan-control/


Good find! It's not just Turing cards that have the asynchronous fan control as a "feature". I have an EVGA 1080Ti with ICX2 cooling that has separate interfaces for controlling the "front" and "rear" fans. On older drivers, I can only control the rear fan speed. By watching the fans under load, if I increase the rear fan speed to 70% using nvidia-settings, the front fan will stop spinning. At that time I didn't think it was a good situation, so I left the fan control on auto on that card. FYI, I've been controlling heat on my cards by lowering the power limit in watts by using nvidia-smi. It's not an ideal solution but it works for me.

http://stefanocappellini.com/monitor_gpu_nvidia-smi/
ID: 52360 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
biodoc

Send message
Joined: 26 Aug 08
Posts: 183
Credit: 10,085,929,375
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52361 - Posted: 28 Jul 2019, 11:26:48 UTC

ID: 52361 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
rod4x4

Send message
Joined: 4 Aug 14
Posts: 266
Credit: 2,219,935,054
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 52362 - Posted: 28 Jul 2019, 12:03:25 UTC - in response to Message 52361.  

FYI, here's how I set fan speed at login.

https://wiki.archlinux.org/index.php/NVIDIA/Tips_and_tricks#Set_fan_speed_at_login

That is a good wiki and gives me a few thing to try. Best guide I have seen for extracting the EDID.

I've been controlling heat on my cards by lowering the power limit in watts by using nvidia-smi. It's not an ideal solution but it works for me.

I use this method to maximize the GPU efficiency. In my case, I can reduce the power draw by 40%, and only get a reduction in output of 15%. Card runs cooler and quieter, power usage is lower... for only a relatively small loss in output.
ID: 52362 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Keith Myers
Avatar

Send message
Joined: 13 Dec 17
Posts: 1419
Credit: 9,119,446,190
RAC: 731
Level
Tyr
Scientific publications
watwatwatwatwat
Message 52363 - Posted: 28 Jul 2019, 17:13:54 UTC - in response to Message 52359.  

I'd like to figure out how nvidia-settings uses the fan control interface identifier on the Turing cards to differentiate the two interfaces.


Does the below link help? I dont have a Turing based card so cant test this.
https://linustechtips.com/main/topic/1048251-turing-rtx-linux-cli-fan-control/

No, I had already figured all that out on my own when I got my first 2080. I use nvidia-settings in a script to set the fan speeds on all my cards.

When the Python code polls the card interfaces with nvidia-settings, it only picks up a single interface in the Python application. You would also have to change the code for the application window to have more radio buttons for the two interfaces and their control.

I am not a programmer, I just dabble in reading code and somewhat understand the logic. But where in the code this fails on Turing escapes me for now.

Anyone want to try and take a crack at the Python script?
ID: 52363 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Aurum
Avatar

Send message
Joined: 12 Jul 17
Posts: 404
Credit: 17,408,899,587
RAC: 0
Level
Trp
Scientific publications
watwatwat
Message 52374 - Posted: 30 Jul 2019, 16:31:00 UTC - in response to Message 52352.  

I wish I knew how to write a script that would shutdown at 1:00 and power on at 6:00.
Shutdown: https://forums.linuxmint.com/viewtopic.php?t=22251
Power on: it can be configured in the BIOS. You should check your local time, and the BIOS clock, as Linux tends to set GMT (=UTC) in the BIOS, so the wake up time should be set according to this.

Thanks. I installed Gshutdown but it seems only able to handle one time events. I need something that will shut down all computers at 1:00 either M-F if TOU electric rate is the issue or every day if heat is the issue. I've heard of cronjob scripts so maybe that's what I need to learn about.
I've seen the wake on timer line in the BIOS. I'll test drive that too.
ID: 52374 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Keith Myers
Avatar

Send message
Joined: 13 Dec 17
Posts: 1419
Credit: 9,119,446,190
RAC: 731
Level
Tyr
Scientific publications
watwatwatwatwat
Message 52378 - Posted: 1 Aug 2019, 20:23:53 UTC - in response to Message 52374.  

I wish I knew how to write a script that would shutdown at 1:00 and power on at 6:00.
Shutdown: https://forums.linuxmint.com/viewtopic.php?t=22251
Power on: it can be configured in the BIOS. You should check your local time, and the BIOS clock, as Linux tends to set GMT (=UTC) in the BIOS, so the wake up time should be set according to this.

Thanks. I installed Gshutdown but it seems only able to handle one time events. I need something that will shut down all computers at 1:00 either M-F if TOU electric rate is the issue or every day if heat is the issue. I've heard of cronjob scripts so maybe that's what I need to learn about.
I've seen the wake on timer line in the BIOS. I'll test drive that too.

If you have the computer on a UPS, then you can schedule a crontab to shut down the host per your schedule and then bring it back up when directed. The UPS interface is already in place to shut the system down and bring it back up for a power event and recovery.

The setting in the BIOS I use is in the power APM events settings. I set the BIOS to "Last State"
ID: 52378 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2

Message boards : Number crunching : NVidia-Linux Adjustments for heat

©2025 Universitat Pompeu Fabra