running GPU without CPU

Message boards : Number crunching : running GPU without CPU
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Lazydude
Volunteer tester

Send message
Joined: 17 Jan 01
Posts: 45
Credit: 96,158,001
RAC: 136
Sweden
Message 2011935 - Posted: 14 Sep 2019, 12:43:51 UTC - in response to Message 2011929.  
Last modified: 14 Sep 2019, 12:55:38 UTC

boinc uses the fastest device first, if boinc run out of gputasks then boinc starts cputasks.
it looks that you are using
app_info.xml

minimal for this is
app_config.xml

<app_config>
<project_max_concurrent>1</project_max_concurrent>
</app_config>
remark: its two diffrent files
ID: 2011935 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 2011949 - Posted: 14 Sep 2019, 16:06:06 UTC - in response to Message 2011935.  

boinc uses the fastest device first, if boinc run out of gputasks then boinc starts cputasks.
it looks that you are using
app_info.xml

minimal for this is
app_config.xml

<app_config>
<project_max_concurrent>1</project_max_concurrent>
</app_config>
remark: its two diffrent files


Thanks again. Your trick did the job. I can now run boinc without noise during the daylight.
ID: 2011949 · Report as offensive
Profile tullio
Volunteer tester

Send message
Joined: 9 Apr 04
Posts: 8797
Credit: 2,930,782
RAC: 1
Italy
Message 2011952 - Posted: 14 Sep 2019, 16:51:16 UTC
Last modified: 14 Sep 2019, 16:53:54 UTC

I am running two GPU tasks, one in GPUGRID and the other in Einsten@home under Science United. The GPUGRID uses 0.966 CPU plus one NVIDIA GPU, the Einstein@Home uses 0.9 CPU plus one NVIDIA GPU.
Tullio
ID: 2011952 · Report as offensive
juan BFP Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 16 Mar 07
Posts: 9786
Credit: 572,710,851
RAC: 3,799
Panama
Message 2011956 - Posted: 14 Sep 2019, 18:52:49 UTC - in response to Message 2011949.  

boinc uses the fastest device first, if boinc run out of gputasks then boinc starts cputasks.
it looks that you are using
app_info.xml

minimal for this is
app_config.xml

<app_config>
<project_max_concurrent>1</project_max_concurrent>
</app_config>
remark: its two diffrent files


Thanks again. Your trick did the job. I can now run boinc without noise during the daylight.

Just an info...

If you run the spoofed client on a host with project_max_concurrent if you CPU cache runs dry you host could not ask for new CPU job.

If that happening stop using project_max_concurrent and control the cpu usage with the cpu usage tag on computing preferences of the host.
ID: 2011956 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 2011969 - Posted: 14 Sep 2019, 21:36:27 UTC - in response to Message 2011956.  

boinc uses the fastest device first, if boinc run out of gputasks then boinc starts cputasks.
it looks that you are using
app_info.xml

minimal for this is
app_config.xml

<app_config>
<project_max_concurrent>1</project_max_concurrent>
</app_config>
remark: its two diffrent files


Thanks again. Your trick did the job. I can now run boinc without noise during the daylight.

Just an info...

If you run the spoofed client on a host with project_max_concurrent if you CPU cache runs dry you host could not ask for new CPU job.

If that happening stop using project_max_concurrent and control the cpu usage with the cpu usage tag on computing preferences of the host.


Not a problem, it take around 1 day to empty the CPU cache (100 CPU tasks with 1 task/hour * number of proc = 25 hours for 4 proc).
More than a typical outage....
ID: 2011969 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14659
Credit: 200,643,578
RAC: 874
United Kingdom
Message 2011972 - Posted: 14 Sep 2019, 22:07:04 UTC - in response to Message 2011969.  

Just an info...

If you run the spoofed client on a host with project_max_concurrent if you CPU cache runs dry you host could not ask for new CPU job.

If that happening stop using project_max_concurrent and control the cpu usage with the cpu usage tag on computing preferences of the host.
Not a problem, it take around 1 day to empty the CPU cache (100 CPU tasks with 1 task/hour * number of proc = 25 hours for 4 proc).
More than a typical outage....
Juan wasn't referring to running dry during outages: he was referring to a known problem where older clients (v7.14.2 and before) might refuse to fetch CPU work if they ever run dry.

Keith and I have worked with David Anderson, and we believe we've got that problem licked with v7.16.1 - we're running tests on the v7.16.1 spoofed client, and it seems to be OK. You do need to be careful not to declare that your GPU is going to need more CPU support than it really needs: my installation is declaring 0.01 CPU to support CUDA 10, and BOINC is keeping the CPUs supplied as I want.
ID: 2011972 · Report as offensive
W3Perl Project Donor
Volunteer tester

Send message
Joined: 29 Apr 99
Posts: 251
Credit: 3,696,783,867
RAC: 12,606
France
Message 2012017 - Posted: 15 Sep 2019, 9:30:01 UTC - in response to Message 2011972.  

Juan wasn't referring to running dry during outages: he was referring to a known problem where older clients (v7.14.2 and before) might refuse to fetch CPU work if they ever run dry.

Keith and I have worked with David Anderson, and we believe we've got that problem licked with v7.16.1 - we're running tests on the v7.16.1 spoofed client, and it seems to be OK. You do need to be careful not to declare that your GPU is going to need more CPU support than it really needs: my installation is declaring 0.01 CPU to support CUDA 10, and BOINC is keeping the CPUs supplied as I want.


Thanks...time to switch to 7.16.1 ;)
ID: 2012017 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : running GPU without CPU


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.