Advanced search

Message boards : Number crunching : Pausing 1 gpu when you have more than 1 in the system.

Author Message
Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 38909 - Posted: 15 Nov 2014 | 2:34:27 UTC

Hi all. been away for a while, but now I'm back. Quick question.
I have 2 GPUs in my system, and I like to crunch on both when possible.
There never used to be a way of pausing only 1 GPU and letting the other one crunch.

Is this still the case or have they fixed that issue?

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 38911 - Posted: 15 Nov 2014 | 8:34:38 UTC - in response to Message 38909.
Last modified: 15 Nov 2014 | 8:37:48 UTC

Well now I just discovered something interesting.
If you go to the Simple View in BOINC Manager (View -> simple View)
You can select a task a suspend it individually by selecting
Task Commands -> Suspend
This is most useful, now I can game/watch a movie etc without having to pause
crunching on both gpus.

Excellent!
if only this option was available in the 'Advanced view" :/

<edit>
You have to check in the advanced view how long the task you want to pause has to go/has run. The simple view doesn't show task names, just a progress bar for each GPUGrid task, so you need that to know which task to pause.

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 0
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38913 - Posted: 15 Nov 2014 | 11:42:04 UTC - in response to Message 38911.
Last modified: 15 Nov 2014 | 12:22:37 UTC

Well now I just discovered something interesting.
If you go to the Simple View in BOINC Manager (View -> simple View)
You can select a task a suspend it individually by selecting
Task Commands -> Suspend
This is most useful, now I can game/watch a movie etc without having to pause
crunching on both gpus.

Excellent!
if only this option was available in the 'Advanced view" :/

It is available in the Advanced view:


This method has a drawback: if you forgot to resume the task, it gets stuck, and the BOINC manager won't ask for new GPUGrid tasks.
To avoid this, you can make two cc_config.xml files, one for using both GPUs, and one for using only one, and two shortcuts to your desktop for copying the desired cc_config.xml to the BOINC manager's folder.

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 38914 - Posted: 15 Nov 2014 | 13:49:50 UTC - in response to Message 38913.

Seems I need to update my BOINC client. I'm still running an older version - 7.2.39

mikey
Send message
Joined: 2 Jan 09
Posts: 297
Credit: 5,833,311,115
RAC: 31,175,033
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38925 - Posted: 16 Nov 2014 | 12:25:01 UTC - in response to Message 38914.
Last modified: 16 Nov 2014 | 12:25:41 UTC

Seems I need to update my BOINC client. I'm still running an older version - 7.2.39


7.4.27 is the latest 'release' version, for Windows, but if you are using a Resource share of zero at any project do NOT upgrade to it as it has problems with that. Other that that I have not found a problem with it, and it DOES show Retvari's feature.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38936 - Posted: 17 Nov 2014 | 3:05:59 UTC - in response to Message 38925.

7.4.27 is the latest 'release' version, for Windows, but if you are using a Resource share of zero at any project do NOT upgrade to it as it has problems with that. Other that that I have not found a problem with it, and it DOES show Retvari's feature.

What's the problem with resource share 0?

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 38939 - Posted: 17 Nov 2014 | 7:04:51 UTC - in response to Message 38925.

Seems I need to update my BOINC client. I'm still running an older version - 7.2.39


7.4.27 is the latest 'release' version, for Windows, but if you are using a Resource share of zero at any project do NOT upgrade to it as it has problems with that. Other that that I have not found a problem with it, and it DOES show Retvari's feature.



All good. I just wanted to be able to pause 1 of the GPUs for a hour or two to have a game, without having to pause both of them.

working well for me!

mikey
Send message
Joined: 2 Jan 09
Posts: 297
Credit: 5,833,311,115
RAC: 31,175,033
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38940 - Posted: 17 Nov 2014 | 13:14:59 UTC - in response to Message 38936.

7.4.27 is the latest 'release' version, for Windows, but if you are using a Resource share of zero at any project do NOT upgrade to it as it has problems with that. Other that that I have not found a problem with it, and it DOES show Retvari's feature.


What's the problem with resource share 0?


It does NOT get any work from that project, even if all other projects don't have any units to send you. The Developers are aware of the problem and are working on a fix. An example of what people are seeing is:
15-Nov-2014 00:17:11 [Asteroids@home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Milkyway@Home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Einstein@Home] [work_fetch] share 0.000 blocked by project preferences

Apparently it sees a zero as never get any work from that place, as opposed to the old way of just keeping a zero cache level, and only getting enough units to crunch right now.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38943 - Posted: 17 Nov 2014 | 15:02:44 UTC - in response to Message 38940.

What's the problem with resource share 0?

It does NOT get any work from that project, even if all other projects don't have any units to send you. The Developers are aware of the problem and are working on a fix. An example of what people are seeing is:
15-Nov-2014 00:17:11 [Asteroids@home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Milkyway@Home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Einstein@Home] [work_fetch] share 0.000 blocked by project preferences

Apparently it sees a zero as never get any work from that place, as opposed to the old way of just keeping a zero cache level, and only getting enough units to crunch right now.

Thanks for the info. Just tried it on an ATI project. Sure enough, didn't work. When I manually polled the project set at zero, I got:

Einstein@Home 11-17-14 08:53 update requested by user
Einstein@Home 11-17-14 08:53 Sending scheduler request: Requested by user.
Einstein@Home 11-17-14 08:53 Not requesting tasks: don't need (CPU: job cache full; NVIDIA GPU: job cache full; AMD/ATI GPU: job cache full)
Einstein@Home 11-17-14 08:53 Scheduler request completed

As long as they're working on this maybe I'll ask for a further addition. It would be useful to have another setting (for instance 1) that would keep a small amount of work in the queue, perhaps 1 hour's worth. This new setting for instance would be useful here for cards that are bumping up against the 24 hour limit.

mikey
Send message
Joined: 2 Jan 09
Posts: 297
Credit: 5,833,311,115
RAC: 31,175,033
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38953 - Posted: 18 Nov 2014 | 14:30:58 UTC - in response to Message 38943.

What's the problem with resource share 0?

It does NOT get any work from that project, even if all other projects don't have any units to send you. The Developers are aware of the problem and are working on a fix. An example of what people are seeing is:
15-Nov-2014 00:17:11 [Asteroids@home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Milkyway@Home] [work_fetch] share 0.000 zero resource share
15-Nov-2014 00:17:11 [Einstein@Home] [work_fetch] share 0.000 blocked by project preferences

Apparently it sees a zero as never get any work from that place, as opposed to the old way of just keeping a zero cache level, and only getting enough units to crunch right now.

Thanks for the info. Just tried it on an ATI project. Sure enough, didn't work. When I manually polled the project set at zero, I got:

Einstein@Home 11-17-14 08:53 update requested by user
Einstein@Home 11-17-14 08:53 Sending scheduler request: Requested by user.
Einstein@Home 11-17-14 08:53 Not requesting tasks: don't need (CPU: job cache full; NVIDIA GPU: job cache full; AMD/ATI GPU: job cache full)
Einstein@Home 11-17-14 08:53 Scheduler request completed

As long as they're working on this maybe I'll ask for a further addition. It would be useful to have another setting (for instance 1) that would keep a small amount of work in the queue, perhaps 1 hour's worth. This new setting for instance would be useful here for cards that are bumping up against the 24 hour limit.


Wouldn't that be more of a 'work buffer' setting in the Boinc Manager than an actual resource share setting? But yes I too dislike the percentage settings there, they are meaningless when running multiple projects and don't apply well when running both cpu and gpu projects, especially when they are not the same one. I think I would prefer a separate setting for each, and even multiple ones if someone has multiple gpu's in the machine. Thru the exclude line one can put each gpu on a separate project, making a single setting a joke.

Of course when they finally give us fine tooth control over each cpu core, they will need a setting for each of those too.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 38954 - Posted: 18 Nov 2014 | 16:25:16 UTC - in response to Message 38953.

As long as they're working on this maybe I'll ask for a further addition. It would be useful to have another setting (for instance 1) that would keep a small amount of work in the queue, perhaps 1 hour's worth. This new setting for instance would be useful here for cards that are bumping up against the 24 hour limit.

Wouldn't that be more of a 'work buffer' setting in the Boinc Manager than an actual resource share setting?

It would allow running a normal work buffer while dealing with the needs of projects like this that need a fast turn around time. It would also be helpful for projects with very small WUs, projects with small WUs combined with large backoff times and projects that have large WU UL/DL sizes. I've run into all these scenarios and it would be beneficial to have such an option to address the issue in addition to the zero-share setting. I sent the above to what I hope was an appropriate thread in the alpha list. Not sure if it's something that would be easily implemented though, and of course not sure if they'll want to do it anyway.

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 39047 - Posted: 30 Nov 2014 | 4:36:05 UTC

I have discovered something mildly interesting with this.

you are running 2 units on 2 cards

-> you pause both units

-> you then resume only 1 of the units

That workunit will resume on the primary GPU,
even if before it was paused it was running on the secondary.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 39050 - Posted: 30 Nov 2014 | 5:57:58 UTC - in response to Message 39047.

This will happen in every BOINC project as the primary GPU always resumes first.

Simba123
Send message
Joined: 5 Dec 11
Posts: 147
Credit: 69,970,684
RAC: 0
Level
Thr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwat
Message 39065 - Posted: 3 Dec 2014 | 13:01:17 UTC - in response to Message 39050.

This will happen in every BOINC project as the primary GPU always resumes first.


Oh, OK. GPUGrid is the only BOINC GPU project I run. I never new that it paused/resumed like that.

Good to know.

Post to thread

Message boards : Number crunching : Pausing 1 gpu when you have more than 1 in the system.

//