Task status has been on "GPU suspended" for days why isn't it using the CPU?

Questions and Answers : GPU applications : Task status has been on "GPU suspended" for days why isn't it using the CPU?
Message board moderation

To post messages, you must log in.

AuthorMessage
Michael Fay

Send message
Joined: 1 Mar 19
Posts: 3
Credit: 227,918
RAC: 0
United States
Message 1984240 - Posted: 9 Mar 2019, 3:26:26 UTC

I'm running a few projects. World Community Grid, Climate Prediction, Rosetta and SETI. All the other tasks for these projects are running but SETI always remains on "GPU Suspended - computer is in use (0.496 CPUS ...")

I'm not sure if I have SETI to only use the GPU (didn't think that was possible) but it doesn't appear to want to run unless the GPU is free.

So SETI is unable to compute and I do not know why. Any help would be appreciated.
ID: 1984240 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1984241 - Posted: 9 Mar 2019, 3:39:11 UTC - in response to Message 1984240.  
Last modified: 9 Mar 2019, 4:04:58 UTC

So SETI is unable to compute and I do not know why. Any help would be appreciated.

You may need to enable Work Fetch Debug for the Event Log to see what is going on, but most likely it's a result of your cache settings, system performance, system up time, time BOINC is actually running & able to do work, and deadlines.

Powerful CPU & GPU systems, that run 24/7, with GPU processing always possible, and CPU processing regardless of non BOINC CPU load can run run multiple projects with large caches without much of an impact on projects such as Seti with long return deadlines.
Low powered systems, with limits on when work can be done & that don't crunch 24/7 and are running more than 1 project are best with very, very small cache settings. The bigger the cache setting and the less time the system processes work each day, then the bigger the impact of short deadline projects on work being done for projects with long deadlines is.
Generally, over time, things (usually) sort themselves out, but it can take (well) over a month- and that's if people don't micro manage.


EDIT- it helps to read the question

"GPU Suspended - computer is in use (0.496 CPUS ...")

It's not running, because your settings are set so it can't run when the computer is in use.

There are web based settings- in your Account, click on "Computing preferences" and make sure
"Suspend when computer is in use"
and
"Suspend GPU computing when computer is in use"
are not selected.

It's also worth setting
"Suspend when non-BOINC CPU usage is above"
to something like 85% if you feel the need to suspend crunching while doing other things on the computer (I don't bother- Seti backs off on it's CPU use as the computer does other things good enough for me).
After changing & saving the changes, click on Update on your computer's BOINC manager to get those changes & have them take effect.


There are also local settings on the computer using the BOINC manager which override any web based settings.
Options, Computing preferences and make sure
"Suspend GPU computing when computer is in use" is not selected. Click on OK to make the changes take effect.
Grant
Darwin NT
ID: 1984241 · Report as offensive
Michael Fay

Send message
Joined: 1 Mar 19
Posts: 3
Credit: 227,918
RAC: 0
United States
Message 1984341 - Posted: 10 Mar 2019, 3:25:59 UTC - in response to Message 1984241.  
Last modified: 10 Mar 2019, 3:26:54 UTC

After watching the logs in real time I do see it say "resuming computation". It looks like it's using the CPU to compute when it says "GPU suspended". I wasn't sure if it was computing because it doesn't say "running" in the status. And "show graphics" button in BOINC is unclickable.

Is this anything holding back computation? the last line about "no projects chosen for work fetch"?

3/9/2019 10:22:34 PM | | [work_fetch] ------- start work fetch state -------
3/9/2019 10:22:34 PM | | [work_fetch] target work buffer: 25920.00 + 43200.00 sec
3/9/2019 10:22:34 PM | | [work_fetch] --- project states ---
3/9/2019 10:22:34 PM | SETI@home | [work_fetch] REC 26064.413 prio -56.151 can't request work: scheduler RPC backoff (117.30 sec)
3/9/2019 10:22:34 PM | | [work_fetch] --- state for CPU ---
3/9/2019 10:22:34 PM | | [work_fetch] shortfall 49384.60 nidle 0.00 saturated 40720.84 busy 0.00
3/9/2019 10:22:34 PM | SETI@home | [work_fetch] share 0.000
3/9/2019 10:22:34 PM | | [work_fetch] --- state for NVIDIA GPU ---
3/9/2019 10:22:34 PM | | [work_fetch] shortfall 0.00 nidle 0.00 saturated 70284.84 busy 0.00
3/9/2019 10:22:34 PM | SETI@home | [work_fetch] share 0.000
3/9/2019 10:22:34 PM | | [work_fetch] ------- end work fetch state -------
3/9/2019 10:22:34 PM | | [work_fetch] No project chosen for work fetch
ID: 1984341 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1984344 - Posted: 10 Mar 2019, 3:57:08 UTC - in response to Message 1984341.  
Last modified: 10 Mar 2019, 4:00:26 UTC

After watching the logs in real time I do see it say "resuming computation". It looks like it's using the CPU to compute when it says "GPU suspended". I wasn't sure if it was computing because it doesn't say "running" in the status.

Work allocated to the CPU can only be processed by the CPU. Work allocated to the GPU can only be processed by the GPU.
If you have the system set to not process GPU work when the system is in use, it won't be processed while the system is in use.

As things stand at present, you have plenty of Seti work to be processed by your GPU, if you change you settings as I posted previously to allow GPU crunching when the system is in use, it will be able to process the work you presently have without delay. With the settings you presently have, it can't process work while you are using your computer.


EDIT- in use means any mouse or keyboard use in the last 3 minutes
ie, it will be at least 3 minutes after any mouse or keyboard use before GPU processing will resume. If you use the mouse or keyboard in that time, the counter is reset again to 3 minutes before processing can begin.
Grant
Darwin NT
ID: 1984344 · Report as offensive
Michael Fay

Send message
Joined: 1 Mar 19
Posts: 3
Credit: 227,918
RAC: 0
United States
Message 1984346 - Posted: 10 Mar 2019, 4:04:56 UTC - in response to Message 1984344.  

Why is some work specifically allocated to GPU computing? Seems to me that a processor is a processor. Why doesn't the CPU take some of that work if the GPU is busy?
ID: 1984346 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13736
Credit: 208,696,464
RAC: 304
Australia
Message 1984349 - Posted: 10 Mar 2019, 4:37:48 UTC - in response to Message 1984346.  
Last modified: 10 Mar 2019, 4:40:12 UTC

Why is some work specifically allocated to GPU computing? Seems to me that a processor is a processor. Why doesn't the CPU take some of that work if the GPU is busy?

Keep in mind the GPU isn't busy- you just have things set so that if you are doing things on the computer, the GPU processing is suspended.

When BOINC was first developed, there were no GPUs to crunch on, it was CPUs only.
And things get complicated because a GPU can process a WU that might take 4 hours on a CPU in as little as a couple of minutes. Also some projects only make use of the CPU, others only the GPU, and some such as Seti can make use of both.
So in order to honour people's cache settings, and project share settings, without missing deadlines on when work is meant to be returned- once work is allocated to a particular processing resource, that's the one that has to process it.
Grant
Darwin NT
ID: 1984349 · Report as offensive

Questions and Answers : GPU applications : Task status has been on "GPU suspended" for days why isn't it using the CPU?


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.