1 Task per GPU

Message boards : Questions and problems : 1 Task per GPU
Message board moderation

To post messages, you must log in.

AuthorMessage
LiLChris

Send message
Joined: 28 Sep 10
Posts: 4
United States
Message 35002 - Posted: 28 Sep 2010, 22:33:59 UTC
Last modified: 28 Sep 2010, 22:37:11 UTC

Ok so I have spent a few days trying to figure it out but have no luck.
What I thought of doing was running 2 instances of BOINC clients with seperate directories.


1st Client dedicating 1st slotted GPU towards Collatz.
<cc_config>
<options>
<data_dir>FIRST CLIENT LOCATION</data_dir>
<use_all_gpus>1</use_all_gpus>
<ignore_cuda_dev>1</ignore_cuda_dev>
</options>
</cc_config>


2nd Client dedicating 2nd slotted GPU towards DNETC.
<cc_config>
<options>
<data_dir>SECOND CLIENT LOCATION</data_dir>
<use_all_gpus>1</use_all_gpus>
<ignore_cuda_dev>0</ignore_cuda_dev>
</options>
</cc_config>



How would I get something like this to work?

Also looked into changing the "app_info.xml" but nothing worked the way I wanted to.

Is there an easier method, if so please point me to the right direction.

Thank you. :)
ID: 35002 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 20 Dec 07
Posts: 1069
Germany
Message 35004 - Posted: 29 Sep 2010, 8:04:02 UTC - in response to Message 35002.  

How did you separate the registry entries for the two clients? ;-)

Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
ID: 35004 · Report as offensive
LiLChris

Send message
Joined: 28 Sep 10
Posts: 4
United States
Message 35005 - Posted: 29 Sep 2010, 9:21:34 UTC - in response to Message 35004.  

I didn't thats why I am asking if there is a way to do it, hence this being a question. :p

Is there a method to dedicate 1 task per GPU?

Specifically DNETC likes to use 2 GPUs for 1 tasks, but there is other reasons why I am looking for a solution to this.

I'm open to suggestions.

ID: 35005 · Report as offensive
skgiven
Avatar

Send message
Joined: 19 Aug 08
Posts: 87
United Kingdom
Message 35159 - Posted: 9 Oct 2010, 19:20:47 UTC - in response to Message 35005.  
Last modified: 9 Oct 2010, 19:21:47 UTC

Hi, is this cc_confiig.xml file sufficient to prevent GPU 0 from being used?

<cc_config>
<options>
<use_all_gpus>0</use_all_gpus>
<ignore_cuda_dev>0</ignore_cuda_dev>
</options>
</cc_config>

XPx86, i7-920, boinc 6.10.58, two GTX470's.

Thanks,
ID: 35159 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 20 Dec 07
Posts: 1069
Germany
Message 35160 - Posted: 9 Oct 2010, 19:56:26 UTC - in response to Message 35159.  

Could be, but you can make sure by looking at your Messages tab. There most configuration items are listed at startup time.

I would suggest <use_all_gpus>1</use_all_gpus> though.

Gruß,
Gundolf
ID: 35160 · Report as offensive
skgiven
Avatar

Send message
Joined: 19 Aug 08
Posts: 87
United Kingdom
Message 35162 - Posted: 9 Oct 2010, 20:42:48 UTC - in response to Message 35160.  
Last modified: 9 Oct 2010, 20:46:31 UTC

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
<ignore_cuda_dev>0</ignore_cuda_dev>
</options>
</cc_config>

09/10/2010 21:32:13 Re-reading cc_config.xml
09/10/2010 21:32:13 Re-read config file
09/10/2010 21:32:13 Config: report completed tasks immediately
09/10/2010 21:32:13 Config: use all coprocessors
09/10/2010 21:32:13 log flags: file_xfer, sched_ops, task

boinc Exit and Start,

09/10/2010 21:33:25 Starting BOINC client version 6.10.58 for windows_intelx86
09/10/2010 21:33:25 Config: report completed tasks immediately
09/10/2010 21:33:25 Config: use all coprocessors
09/10/2010 21:33:25 log flags: file_xfer, sched_ops, task
09/10/2010 21:33:25 Libraries: libcurl/7.19.7 OpenSSL/0.9.8l zlib/1.2.3

After both a read and a restart nothing changed; still using 2 GPUs. Per chance, do existing tasks have to complete first?

- Or even, would I need to remove card 0 from the system, reboot, start Boinc and then shut down, install the card, reboot and start Boinc again?

Thanks,
ID: 35162 · Report as offensive
skgiven
Avatar

Send message
Joined: 19 Aug 08
Posts: 87
United Kingdom
Message 35163 - Posted: 9 Oct 2010, 21:55:27 UTC - in response to Message 35162.  

OK, looks like running and queued tasks need to finish/get terminated, a project detatch or reset, several restarts and so on need to be done for this to work.
ID: 35163 · Report as offensive
Profile Gundolf Jahn

Send message
Joined: 20 Dec 07
Posts: 1069
Germany
Message 35164 - Posted: 9 Oct 2010, 22:47:39 UTC - in response to Message 35163.  

What about the log entries for the GPUs? Maybe one is marked as "not used".

Gruß,
Gundolf
ID: 35164 · Report as offensive
skgiven
Avatar

Send message
Joined: 19 Aug 08
Posts: 87
United Kingdom
Message 35173 - Posted: 10 Oct 2010, 17:29:01 UTC - in response to Message 35164.  

Message reads as follows,
Config: ignoring NVIDIA GPU 0
ID: 35173 · Report as offensive

Message boards : Questions and problems : 1 Task per GPU

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.