Advanced search

Message boards : Number crunching : Does anyone have an app_config.xml that works?

Author Message
Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31080 - Posted: 27 Jun 2013 | 22:00:45 UTC

Does anyone have an app_config.xml that works for GPUGRID? I'm running multiple machines with dual GTX690's, dual GTX590's, and two mixed machines with GTX690/660Ti, and GTX590/GTX660Ti. I would like to experiment with .5/gpu configurations to see if I could squeeze more work out of all this computing power. Thanks in advance to anyone/everyone who responds. Rick
____________

captainjack
Send message
Joined: 9 May 13
Posts: 171
Credit: 3,570,489,156
RAC: 18,299,904
Level
Arg
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31081 - Posted: 28 Jun 2013 | 0:50:27 UTC

Try this one:

http://www.gpugrid.net/forum_thread.php?id=3359&nowrap=true#31014

Hope that helps.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31112 - Posted: 28 Jun 2013 | 19:43:18 UTC - in response to Message 31080.

I would like to experiment with .5/gpu configurations to see if I could squeeze more work out of all this computing power.

Hi Rick. We've been through this extensively. Short answer: not a good idea.

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31115 - Posted: 28 Jun 2013 | 20:58:54 UTC

See this post just from this morning - doesn't seem like a good idea on typical long-run WUs. And you wouldn't want to run short-runs on this hardware, as long as any long-runs are left.

MrS
____________
Scanning for our furry friends since Jan 2002

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31121 - Posted: 29 Jun 2013 | 1:56:01 UTC - in response to Message 31081.
Last modified: 29 Jun 2013 | 1:57:42 UTC

Try this one:

http://www.gpugrid.net/forum_thread.php?id=3359&nowrap=true#31014

Hope that helps.

Thanks for the help CaptinJack, and the comments from the rest of you. I tried the app_config from your link and it does work. Based on your comments,rather than try to run multiple GPUGRIDs per processor, I'm trying to run grugrid ~ .7, and Other projects such as seti, Einstein, and Milkyway at ~ .3. Well see if this works, but with the first 24Hrs. under my belt, it seems to be worth the experiment. Rick
____________

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31123 - Posted: 29 Jun 2013 | 9:21:20 UTC - in response to Message 31121.

Don't overburden the GPU with other apps or the time it takes the GPUGrid WU will increase dramatically. Going by GPU usage is a bad indicator, go by runtime, tasks returned/day and credit. Overall you might return more work or get better GPU credit running 2 tasks (1 GPUGrid and 1 at another light project), but when I tried more than 1 WU from another project the GPUGrid WU times increased massively (two or three times as long). You should also note that if you over-commit the CPU, GPU performance may be reduced exponentially. If for example you wanted to run POEM WU's + GPUGrid WU's on the same card don't use any more than 50% of the CPU cores. If you start seeing failures, all gains will be lost.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31131 - Posted: 29 Jun 2013 | 21:01:34 UTC - in response to Message 31123.
Last modified: 29 Jun 2013 | 21:03:04 UTC

I can confirm; a 30% reduction in GPU core allocation (from 1.0 to 0.7) results in a 100% increase in time required to complete a GPUGRID WU. Basically the same results on my GTX590's, GTX660Ti's, and GTX690's. I can't explain why, but that's what happened in my tests. Rick
____________

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31225 - Posted: 4 Jul 2013 | 20:32:40 UTC - in response to Message 31131.
Last modified: 4 Jul 2013 | 20:32:52 UTC

Hi Rick,

what exactly did you run along GPU-Grid set to 0.7 GPU? Note that this value is only used to help the BOINC scheduler decide which WUs to launch. It does not affect the actual GPU usage. So if you run e.g. GPU-Grid at 0.7 and Einstein at 0.3, they'll probably share the GPU 0.5 : 0.5. Or each might get as many time slices as the other one, but their length would depend on the WU and app, so sharing would be asynchronous in some hard-to-predict way.

Hence just looking at the GPU-Grid time is only half of the story. If the throughput of the other project increased, it might still be worth it.

MrS
____________
Scanning for our furry friends since Jan 2002

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31257 - Posted: 5 Jul 2013 | 16:15:43 UTC - in response to Message 31225.

Hi Rick,

what exactly did you run along GPU-Grid set to 0.7 GPU? Note that this value is only used to help the BOINC scheduler decide which WUs to launch. It does not affect the actual GPU usage. So if you run e.g. GPU-Grid at 0.7 and Einstein at 0.3, they'll probably share the GPU 0.5 : 0.5. Or each might get as many time slices as the other one, but their length would depend on the WU and app, so sharing would be asynchronous in some hard-to-predict way.

Hence just looking at the GPU-Grid time is only half of the story. If the throughput of the other project increased, it might still be worth it.

MrS

Wow, I was not expecting that behavior (50/50) in BOINC with a .7 GPU allocation to GPUGrid. I've run GUOPGrid with , Seti, Einstein, and Milkyway. All these other projects seem to benefit from sharing a GPU, that's why I was trying the same for GPUGrid. I'll continue to experiment, but it doen't look good. Thanks for the explanation MrS. Regards, Rick
____________

ExtraTerrestrial Apes
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 17 Aug 08
Posts: 2705
Credit: 1,311,122,549
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 31266 - Posted: 5 Jul 2013 | 20:28:57 UTC - in response to Message 31257.

Actually I suspect this 50/50 only applies to running 2 similar WUs at once. Even for different WUs from GPU-Grid it probably differs.

The problem is that currently not even the OS could enforce GPU usage. A thread is allowed to send its instructions to the GPU, the GPU processes them, sends the results back.. and only then becomes available for other work. That's why the display becomes choppy under heavy load; when these individual "work chunks" take too long. This is the "time per step" reported by GPU-Grid.

And when not even the OS has proper scheduling and multitasking for GPUs yet, what could BOINC do about this :/

I suppose mixing one GPU-Grid and one Einstein will be fine, as GPU utilization is significantly higher this way. Although the new Nathan KIXors kick so much ***, eh GPU, that running anything along can not possibly help.

MrS
____________
Scanning for our furry friends since Jan 2002

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32407 - Posted: 28 Aug 2013 | 19:18:21 UTC

Ok, I'm getting wu's again, BUT I give up on what new app names to use in the app_config.xml file. Can anyone help me? Thanks in advance, Rick
____________

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,206,655,749
RAC: 261,147
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32424 - Posted: 28 Aug 2013 | 22:12:59 UTC - in response to Message 32407.

Ok, I'm getting wu's again, BUT I give up on what new app names to use in the app_config.xml file. Can anyone help me? Thanks in advance, Rick

acemd.800-55.exe

GDB
Send message
Joined: 24 Oct 11
Posts: 4
Credit: 385,332,304
RAC: 1,454,625
Level
Asp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32431 - Posted: 29 Aug 2013 | 0:17:43 UTC - in response to Message 32424.

You can get the app_name by looking in the BOINC Manager Event Log for "Starting task" events. Following "using" is app_name.

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32508 - Posted: 29 Aug 2013 | 21:11:20 UTC - in response to Message 32431.

That was what I thought, but when I put "acemd.800-55.exe" or "acemd.800-55" or "acemd_800_55" and about 10 other variations, BOINC did not recognize the app name. Couriously, the new long app seems to accept the variables of the old "acemdlong" app name in my app_config.xml file. Rick
____________

Profile nenym
Send message
Joined: 31 Mar 09
Posts: 137
Credit: 1,308,230,581
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32509 - Posted: 29 Aug 2013 | 21:23:23 UTC - in response to Message 32508.
Last modified: 29 Aug 2013 | 21:25:29 UTC

You must use <name> of <app> section (=acemdlong, acemdbeta), not <name> of application.
http://boinc.berkeley.edu/trac/wiki/ClientAppConfig
If you want to change anything depending on application name,you must use app_info.xml.

Jacob Klein
Send message
Joined: 11 Oct 08
Posts: 1127
Credit: 1,901,927,545
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32589 - Posted: 1 Sep 2013 | 12:33:41 UTC - in response to Message 32508.

That was what I thought, but when I put "acemd.800-55.exe" or "acemd.800-55" or "acemd_800_55" and about 10 other variations, BOINC did not recognize the app name. Couriously, the new long app seems to accept the variables of the old "acemdlong" app name in my app_config.xml file. Rick


I get the app names from looking in the client_state.xml file, within the <app> blocks.

Rick A. Sponholz
Avatar
Send message
Joined: 20 Jan 09
Posts: 52
Credit: 2,518,707,115
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32827 - Posted: 8 Sep 2013 | 15:17:01 UTC - in response to Message 32589.

That was what I thought, but when I put "acemd.800-55.exe" or "acemd.800-55" or "acemd_800_55" and about 10 other variations, BOINC did not recognize the app name. Couriously, the new long app seems to accept the variables of the old "acemdlong" app name in my app_config.xml file. Rick


I get the app names from looking in the client_state.xml file, within the <app> blocks.


Thanks Jacob. That's exactly what I was looking for. Regards, Rick
____________

Jacob Klein
Send message
Joined: 11 Oct 08
Posts: 1127
Credit: 1,901,927,545
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 32831 - Posted: 8 Sep 2013 | 16:02:56 UTC - in response to Message 32827.
Last modified: 8 Sep 2013 | 16:04:27 UTC

No problem.

For full disclosure, here is my GPUGrid app_config.xml file (including tons of personal notes in xml comments at the top that briefly describe my testing and results over the course of the last several months)

<!-- GPUGrid.net -->
<!-- GPU tasks do properly use higher process and thread priorities, compared to CPU tasks. -->
<!-- GPU tasks sometimes use CPU sometimes don't, based on type of GPU task runs on. -->
<!-- Recommend 1 gpu_usage, if user also has CPU projects. -->
<!-- Recommend 0.001 cpu_usage, but might try 0.5, since if 2 are running, I KNOW the Kepler is using CPU -->
<!-- Also might try 1 cpu_usage, so as not to overcommit per Task Manager's CPU Utilization -->
<!-- Although x-at-a-time provides the best per-task-throughput, it ends up using a lot more CPU -->
<!-- Switching to 0.4995, such that if an 8-CPU MT job is running, 2 GPUGrid jobs and 1 0.001 GPU job can all run together -->
<!-- 0.5 cpu_usage so that 2+ GPU tasks will intentionally reserve at least 1 core -->
<!-- 1.0 cpu_usage because, when SETI tasks run on 3rd GPU reserving a core, they still aren't getting enough CPU -->

<app_config>

<!-- Short runs (2-3 hours on fastest card) -->
<app>
<name>acemdshort</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<!-- Long runs (8-12 hours on fastest card) -->
<app>
<name>acemdlong</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

<!-- ACEMD beta version -->
<app>
<name>acemdbeta</name>
<max_concurrent>0</max_concurrent>
<gpu_versions>
<gpu_usage>1</gpu_usage>
<cpu_usage>1</cpu_usage>
</gpu_versions>
</app>

</app_config>

Post to thread

Message boards : Number crunching : Does anyone have an app_config.xml that works?

//