Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Aggie The Pew message board
: How to assign GPU's?
Author |
Message |
Neo Volunteer tester
 Send message
Joined: 28 Oct 10 Posts: 710 ID: 71509 Credit: 91,178,992 RAC: 0
                   
|
Despite Primegrid reporting that I have 2 750GTX gpus....
I really have a 750 and a 770.
I would like to run some genefer tasks on my 770 and let my 750 work on sieving..
Perhaps even give my 750 two sieve tasks at a time...
Gary: What lines do I need to add to my config file to do this?
BOINC is reporting that my 770 is GPU 0
and my 750 is GPU 1
Thanks in advance.
Neo | |
|
|
i want to run two gfns on one card as well so any help would be appreciated. Or even one gfn and one pps sieve. | |
|
|
Here's the content of my app_config.xml file that enables two parallel PPS Sieve tasks on a single (770) card. This file goes in the projects/www.primegrid.com directory:
<app_config>
<app>
<name>pps_sr2sieve</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
</app_config>
For genefer, I'd imagine you'd just need to replace the "pps_sr2sieve" with the genefer app name, which I think is "genefer" or "genefer_wr", but check that in your client_state.xml file. To make settings for both PPS/Sieve and Genefer, I think you'd replicate everything between (and including) the <app> tags, changing the <name> appropriately.
For assigning specific subprojects to specific GPUs, I've never tried that. However, you might try the <exclude_gpu> tag in app_config.xml. Documentation is here: http://boinc.berkeley.edu/wiki/client_configuration. So maybe you could exclude GPU 0 for sieve and exclude GPU 1 for genefer. At least that's the way I read the documentation.
My <max_concurrent> is set to 4 above, so as to allow 2 GPU and 2 CPU tasks to run (that was for the December challenge). Currently my prefs disable CPU sieve work, so just 2 GPU tasks are running.
It is necessary to have a reasonably current version of BOINC installed for this to work. 6.x.x is definitely too old. I think 7.0.40-something is okay.
G | |
|
|
Here's the content of my app_config.xml file that enables two parallel PPS Sieve tasks on a single (770) card. This file goes in the projects/www.primegrid.com directory:
<app_config>
<app>
<name>pps_sr2sieve</name>
<max_concurrent>4</max_concurrent>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
</app_config>
For genefer, I'd imagine you'd just need to replace the "pps_sr2sieve" with the genefer app name, which I think is "genefer" or "genefer_wr", but check that in your client_state.xml file. To make settings for both PPS/Sieve and Genefer, I think you'd replicate everything between (and including) the <app> tags, changing the <name> appropriately.
For assigning specific subprojects to specific GPUs, I've never tried that. However, you might try the <exclude_gpu> tag in app_config.xml. Documentation is here: http://boinc.berkeley.edu/wiki/client_configuration. So maybe you could exclude GPU 0 for sieve and exclude GPU 1 for genefer. At least that's the way I read the documentation.
My <max_concurrent> is set to 4 above, so as to allow 2 GPU and 2 CPU tasks to run (that was for the December challenge). Currently my prefs disable CPU sieve work, so just 2 GPU tasks are running.
It is necessary to have a reasonably current version of BOINC installed for this to work. 6.x.x is definitely too old. I think 7.0.40-something is okay.
G
thanks. Once I get my current goal on pps sieve i plan on testing this | |
|
|
I decided to test now. I think I have it working and have a gfn short and a pps sieve currently running at the same time. I have no idea how this will work as I get more units but here is my app_config.xml file at the moment. I changed max_concurrent to 1 to see if that limits each instance to running just one unit at at time.
Sorry Neo, I don't have multiple cards so I couldn't test that for you.
<app_config>
<app>
<name>pps_sr2sieve</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
<app>
<name>genefer</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>0.1</cpu_usage>
</gpu_versions>
</app>
</app_config>
| |
|
|
Not sure anyone else cares but an update on this dual job test. I was able to get both gfn shorts and pps sieve jobs BUT it's a manual process due to the cache. Basically Boinc still knows I have only 1 card and will only send tasks to cover the cache I have set. So it doesn't replenish each type like normal. I have to manually set the preferences once I need a specific job type. It works but it's monitored.
I got one gfn (ocl) done in 51,529.65 seconds (14.31 hrs) which is not too terribly slow since I'm also running pps sieves. This is on my 660ti card. | |
|
|
Moved my 770 off WFS and back onto GFN shorts. Got 2 shorts running at the same time and at the moment it "says" they will be done in 8.5 hours. Not sure how good or bad that is but it's running 99% of my GPU and 50% of my cpu so I'm going to let it crank out. | |
|
|
Moved my 770 off WFS and back onto GFN shorts. Got 2 shorts running at the same time and at the moment it "says" they will be done in 8.5 hours. Not sure how good or bad that is but it's running 99% of my GPU and 50% of my cpu so I'm going to let it crank out.
That's an excellent run-time for gfn-short. Let them run. A reliable estimate for gfn run-time can be found in the slots/*/stderr.txt file. Much better than the GUI manager estimate. | |
|
Message boards :
Aggie The Pew message board
: How to assign GPU's? |