Getting GPU work when told not to

Message boards : Questions and problems : Getting GPU work when told not to
Message board moderation

To post messages, you must log in.

1 · 2 · Next

AuthorMessage
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106979 - Posted: 3 Feb 2022, 12:48:00 UTC
Last modified: 3 Feb 2022, 12:52:48 UTC

So that setting, "suspend GPU", doesn't seem to bother stopping the download of work [rolls eyes]. I have a machine which does CPU and GPU work and I don't want the GPU running at the moment (saving power). Yet it downloads a queue of GPU work. I guess it's just going to hog work units until they expire then get another batch. What a waste.

Posted on github: https://github.com/BOINC/boinc/issues/4623
ID: 106979 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 4792
United Kingdom
Message 106980 - Posted: 3 Feb 2022, 12:50:43 UTC - in response to Message 106979.  

Which project? Does the Event Log contain lines commencing "Resent lost task ..." for the download event?
ID: 106980 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106981 - Posted: 3 Feb 2022, 12:54:10 UTC - in response to Message 106980.  

Milkyway@Home 03-02-2022 12:43 PM work fetch resumed by user
Milkyway@Home 03-02-2022 12:44 PM Sending scheduler request: To fetch work.
Milkyway@Home 03-02-2022 12:44 PM Requesting new tasks for CPU and AMD/ATI GPU
Milkyway@Home 03-02-2022 12:44 PM Scheduler request completed: got 333 new tasks
Milkyway@Home 03-02-2022 12:44 PM Project requested delay of 91 seconds
Milkyway@Home 03-02-2022 12:44 PM Starting task de_nbody_08_31_2021_v176_40k__data__11_1640968180_2538080_1
ID: 106981 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 4792
United Kingdom
Message 106982 - Posted: 3 Feb 2022, 13:01:33 UTC

Set preference "Don't use GPU" at the project.
ID: 106982 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106983 - Posted: 3 Feb 2022, 13:36:02 UTC - in response to Message 106982.  
Last modified: 3 Feb 2022, 13:37:51 UTC

Yes I've done that as a workaround. But surely it shouldn't download until it's set to run again? It could be off for a day or more. And far more convenient for the user to be able to stop the GPU running with one click instead of going to possibly several project pages to turn it off.

I seem to have been added to number 243 in the backlog.... and number 37 in client/scheduler policy.
ID: 106983 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14993
Netherlands
Message 106984 - Posted: 3 Feb 2022, 13:56:51 UTC - in response to Message 106983.  
Last modified: 3 Feb 2022, 13:57:40 UTC

Yes I've done that as a workaround.
That's not a workaround, that is the way to do it. Appointing hardware resources is a project preference, so you set this preference via the projects web pages, all of them and if that's a bother, you use an account manager.

Vitalii is correct in your ticket #4623 that the local client doesn't know if the GPU is going to be disabled forever or not, that's a decision you make and so you should communicate it correctly to the client.
ID: 106984 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106985 - Posted: 3 Feb 2022, 15:31:18 UTC - in response to Message 106984.  
Last modified: 3 Feb 2022, 15:34:24 UTC

Yes I've done that as a workaround.
That's not a workaround, that is the way to do it. Appointing hardware resources is a project preference, so you set this preference via the projects web pages, all of them and if that's a bother, you use an account manager.

Vitalii is correct in your ticket #4623 that the local client doesn't know if the GPU is going to be disabled forever or not, that's a decision you make and so you should communicate it correctly to the client.
I did communicate it correctly, I said "suspend GPU work". It's illogical to download work since it doesn't know if and when I'll re-enable it. If your boss tells you to take some leave because you look tired, you don't go hunting for more work to add to your pile, causing the clients to have an unknown delay.
ID: 106985 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 4792
United Kingdom
Message 106986 - Posted: 3 Feb 2022, 16:10:20 UTC - in response to Message 106985.  

In your opening post, you say that you are pausing GPU work "to save power". That sounds like a long term decision - not just a five minute tea break while you have a quick game of solitaire.

In which case, you could consider excluding the Milky Way project from your GPU(s) in cc_config.xml
ID: 106986 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106987 - Posted: 3 Feb 2022, 16:15:10 UTC - in response to Message 106986.  

I use the computers to heat the house, so in warmer weather I turn GPUs off.

Also, I've noticed changing the settings at the server end only takes effect the second time contact is made. Presumably the client asks for CPU and GPU work, then is told the new settings by the server. This should be done the other way round.
ID: 106987 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14993
Netherlands
Message 106988 - Posted: 3 Feb 2022, 16:40:52 UTC - in response to Message 106985.  

If your boss tells you to take some leave because you look tired, you don't go hunting for more work to add to your pile, causing the clients to have an unknown delay.
But your boss will still pile up some work for you to do when you come back. Because this is your specialty.

If you do not want your boss (== the project) to pile up work that only you can do, you tell him you're not available (== at project preferences, uncheck use of GPU).
ID: 106988 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 106990 - Posted: 3 Feb 2022, 20:02:03 UTC - in response to Message 106988.  

If your boss tells you to take some leave because you look tired, you don't go hunting for more work to add to your pile, causing the clients to have an unknown delay.
But your boss will still pile up some work for you to do when you come back. Because this is your specialty.

If you do not want your boss (== the project) to pile up work that only you can do, you tell him you're not available (== at project preferences, uncheck use of GPU).
Ah but my Boinc client is not in the position of being the only one that can do it. Boinc is more like a huge group of similarly skilled staff, each with a pile of work on their desk. You don't stick more work on the desk of the guy on leave.
ID: 106990 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 107070 - Posted: 13 Feb 2022, 16:23:33 UTC - in response to Message 106986.  
Last modified: 13 Feb 2022, 16:24:06 UTC

In your opening post, you say that you are pausing GPU work "to save power". That sounds like a long term decision - not just a five minute tea break while you have a quick game of solitaire.

In which case, you could consider excluding the Milky Way project from your GPU(s) in cc_config.xml
Just encountered a similar problem, but for CPUs. If I only allow work from projects which have "no CPU work" set on the server, it will still request it as a last resort. Perhaps that's useful, if the project you want CPU work from runs out temporarily. But what happens with me is I'm in the middle of allowing new work from different projects, and it grabs what I didn't want before I've turned on the other project.
ID: 107070 · Report as offensive
Bryn Mawr
Help desk expert

Send message
Joined: 31 Dec 18
Posts: 229
United Kingdom
Message 107071 - Posted: 13 Feb 2022, 19:32:45 UTC - in response to Message 107070.  

In your opening post, you say that you are pausing GPU work "to save power". That sounds like a long term decision - not just a five minute tea break while you have a quick game of solitaire.

In which case, you could consider excluding the Milky Way project from your GPU(s) in cc_config.xml
Just encountered a similar problem, but for CPUs. If I only allow work from projects which have "no CPU work" set on the server, it will still request it as a last resort. Perhaps that's useful, if the project you want CPU work from runs out temporarily. But what happens with me is I'm in the middle of allowing new work from different projects, and it grabs what I didn't want before I've turned on the other project.


Do you have”If there is no work available for the project(s) I have selected above, please send me work from another project.” (as an example, this from WCG) ticked?
ID: 107071 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 107072 - Posted: 13 Feb 2022, 21:29:20 UTC - in response to Message 107071.  

In your opening post, you say that you are pausing GPU work "to save power". That sounds like a long term decision - not just a five minute tea break while you have a quick game of solitaire.

In which case, you could consider excluding the Milky Way project from your GPU(s) in cc_config.xml
Just encountered a similar problem, but for CPUs. If I only allow work from projects which have "no CPU work" set on the server, it will still request it as a last resort. Perhaps that's useful, if the project you want CPU work from runs out temporarily. But what happens with me is I'm in the middle of allowing new work from different projects, and it grabs what I didn't want before I've turned on the other project.


Do you have”If there is no work available for the project(s) I have selected above, please send me work from another project.” (as an example, this from WCG) ticked?
That could be possible. I wasn't aware that allowed them for other types of processor though. Very odd the way things are set up.
ID: 107072 · Report as offensive
ProDigit

Send message
Joined: 8 Nov 19
Posts: 706
United States
Message 107091 - Posted: 15 Feb 2022, 23:53:36 UTC

I wonder if the suspend option, is only meant as to temporarily disable the GPU.
Hence, the work is downloaded anyway.
Disable the GPU is more of a permanent solution.
But I get what you're saying, even with suspend, no new GPU tasks should be downloaded.
ID: 107091 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 107138 - Posted: 17 Feb 2022, 22:48:33 UTC - in response to Message 107091.  

I wonder if the suspend option, is only meant as to temporarily disable the GPU.
Hence, the work is downloaded anyway.
Disable the GPU is more of a permanent solution.
But I get what you're saying, even with suspend, no new GPU tasks should be downloaded.
It's really annoying me now. I just got a load of Gamma CPU work from Einstein when I wasn't intending to get any CPU work (I had Folding@Home on that processor). I'd set Einstein to GPU only as I wanted to do their Gravity application. Github bug report filled in here: https://github.com/BOINC/boinc/issues/4641
ID: 107138 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 4792
United Kingdom
Message 107142 - Posted: 18 Feb 2022, 8:41:18 UTC - in response to Message 107138.  

From the log you posted at GitHub:

590 Einstein@Home 17-02-2022 10:24 PM [sched_op] Starting scheduler request
591 Einstein@Home 17-02-2022 10:24 PM Sending scheduler request: To fetch work.
592 Einstein@Home 17-02-2022 10:24 PM Requesting new tasks for CPU
593 Einstein@Home 17-02-2022 10:24 PM [sched_op] CPU work request: 757459.98 seconds; 7.00 devices
594 Einstein@Home 17-02-2022 10:24 PM [sched_op] AMD/ATI GPU work request: 0.00 seconds; 0.00 devices
595 Einstein@Home 17-02-2022 10:24 PM Scheduler request completed: got 7 new tasks
You asked, you got. That's how Einstein works.

Did you follow the advice you've been given in #4624? In the context of that issue, Einstein is running a very old server - the base version reports 611, so they won't have the patch that appears to have been applied between 713 and 717
ID: 107142 · Report as offensive
Peter Hucker of the Scottish B...
Avatar

Send message
Joined: 6 Oct 06
Posts: 1491
United Kingdom
Message 107143 - Posted: 18 Feb 2022, 9:16:16 UTC - in response to Message 107142.  

From the log you posted at GitHub:

590 Einstein@Home 17-02-2022 10:24 PM [sched_op] Starting scheduler request
591 Einstein@Home 17-02-2022 10:24 PM Sending scheduler request: To fetch work.
592 Einstein@Home 17-02-2022 10:24 PM Requesting new tasks for CPU
593 Einstein@Home 17-02-2022 10:24 PM [sched_op] CPU work request: 757459.98 seconds; 7.00 devices
594 Einstein@Home 17-02-2022 10:24 PM [sched_op] AMD/ATI GPU work request: 0.00 seconds; 0.00 devices
595 Einstein@Home 17-02-2022 10:24 PM Scheduler request completed: got 7 new tasks
You asked, you got. That's how Einstein works.
I didn't ask. Einstein's server was set to no CPU work, simple as that. It ignored me. My client ignored me, not Einstein, you can see from my log that my own computer is asking for something it shouldn't. It's nothing to do with Einstein's server.

Did you follow the advice you've been given in #4624?
Not sure which advice you want me to follow.

In the context of that issue, Einstein is running a very old server - the base version reports 611, so they won't have the patch that appears to have been applied between 713 and 717
ID: 107143 · Report as offensive
robsmith
Volunteer tester
Help desk expert

Send message
Joined: 25 May 09
Posts: 1101
United Kingdom
Message 107145 - Posted: 18 Feb 2022, 12:17:51 UTC - in response to Message 107143.  

You, or at least your computer requested a pile of CPU work:
591 Einstein@Home 17-02-2022 10:24 PM Sending scheduler request: To fetch work.
592 Einstein@Home 17-02-2022 10:24 PM Requesting new tasks for CPU


It would appear that you are using your local set of preferences and not those held by Einstein, and so you made a call which the server obeyed.

As Richard then pointed out Einstein are running an old version of the server software, and not the current one. BOINC (the project) has no control over what version of code the projects run, so your best line of attack would be to go through the Einstein forum and push them to update to the current server code.
On the other hand, somewhere you say you are using an account manager, are you sure that is doing what you desire?
ID: 107145 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 14993
Netherlands
Message 107146 - Posted: 18 Feb 2022, 12:37:41 UTC

Einstein has some extra Project preferences which may interfere here:

Of course there's: "Run only the selected applications"
With options:
Run only the selected applications
Binary Radio Pulsar Search (Arecibo)
Binary Radio Pulsar Search (Arecibo, GPU)
Gamma-ray pulsar binary search #1
Gamma-ray pulsar search #5
Gamma-ray pulsar binary search #1 (GPU)
Gravitational Wave search O2 Multi-Directional
Gravitational Wave search O2 Multi-Directional GPU
Gravitational Wave search O3 All-Sky
Then there's these:

Run CPU versions of applications for which GPU versions are available:

Allow non-preferred apps:
If no work for selected applications are available, accept tasks from other applications?
If these are set to Yes, BOINC will happily follow through and ask for CPU work even if "Use CPU?" is set to No. Doesn't matter which server version they sport.
As ever, with running BOINC comes some responsibility from the user, to set all the possible preferences correctly. The client and the server don't (necessarily) know about a CPU or GPU, but they do know about the science applications (through the plan class). And neither cares what hardware resource those applications use.
ID: 107146 · Report as offensive
1 · 2 · Next

Message boards : Questions and problems : Getting GPU work when told not to

Copyright © 2022 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.