Advanced search

Message boards : Number crunching : Question on BOINC client config

Author Message
GPUGRID Role account
Send message
Joined: 15 Feb 07
Posts: 134
Credit: 1,349,535,983
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 13894 - Posted: 12 Dec 2009 | 11:36:59 UTC

Hi,

I want to reconfigure the BOINC client to use only 1 GPU (my hosts have 2) and to request only work unit, rather than the project default (2/GPU). Could someone tell me whether that's possible, please?

MJH

MarkJ
Volunteer moderator
Volunteer tester
Send message
Joined: 24 Dec 08
Posts: 738
Credit: 200,909,904
RAC: 0
Level
Leu
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 13897 - Posted: 12 Dec 2009 | 14:16:13 UTC - in response to Message 13894.
Last modified: 12 Dec 2009 | 14:17:43 UTC

Hi,

I want to reconfigure the BOINC client to use only 1 GPU (my hosts have 2) and to request only work unit, rather than the project default (2/GPU). Could someone tell me whether that's possible, please?

MJH


Well you can do the 1st bit by adding <ignore_cuda_dev>n</ignore_cuda_dev> to your cc_config file. Where n is the cuda device number you want it to ignore. You'll need version 6.10.19 or later for this to work.

As to the 2nd part of limiting the number of work units, it seems the quotas override things and you get 2 per gpu regardless. I'm not aware of a work around for this one, although some of the guys have suggested reducing the cache to 0 days.
____________
BOINC blog

Profile Michael Goetz
Avatar
Send message
Joined: 2 Mar 09
Posts: 124
Credit: 7,573,744
RAC: 0
Level
Ser
Scientific publications
watwatwatwatwatwatwatwat
Message 13908 - Posted: 13 Dec 2009 | 13:16:22 UTC - in response to Message 13897.

The only way I know of to limit your computer to only download only one WU at a time is to lower the network settings for the whole computer. I have "Connect about every" set to 0.00 and "Additional work buffer" set to 0.04 (which is approximately 1 hour). The result of these settings should be that it won't load another task until there's only an hour to go on the running task. In practice, it doesn't work quite that way, but it's fairly close.

Set like that, it tends to wait until there's around 30 minutes to go on the running GPUGRID WU and then download the next one.

When running shorter (or more variable) GPU projects such as Milkyway or SETI, it will keep 2 to 4 WUs in the queue since they're so short (minutes).

Of course, this affects ALL projects, including those on the CPU as well, and not just the GPU.

You can also set both numbers to 0. I originally did that, but wasn't pleased with the results for some reason; I don't remember why.

Post to thread

Message boards : Number crunching : Question on BOINC client config

//