CUDA Taking Too Long To Process?

Message boards : Number crunching : CUDA Taking Too Long To Process?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Steve Dodd

Send message
Joined: 29 May 99
Posts: 23
Credit: 8,695,373
RAC: 1
United States
Message 922420 - Posted: 30 Jul 2009, 13:20:17 UTC

I got this message for a 6.08 wu (using 6.6.37 BOINC) -

7/30/2009 3:26:42 AM SETI@home Aborting task 04dc08ah.16808.1299.15.10.131_0: exceeded elapsed time limit 16701.964037

Didn't expect to see this.
XP 32-bit, Q6600, GTX 260 - 216 (driver 182.15)

I've been unable to successfully process any GPUGRID wu, but that's another story. I see the occasional aborted SETI CUDA wu abort, but I caught this one in the messages.

Any info.? Is this related to the VLAR issue?
ID: 922420 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 922432 - Posted: 30 Jul 2009, 14:33:20 UTC - in response to Message 922420.  

I got this message for a 6.08 wu (using 6.6.37 BOINC) -

7/30/2009 3:26:42 AM SETI@home Aborting task 04dc08ah.16808.1299.15.10.131_0: exceeded elapsed time limit 16701.964037

Didn't expect to see this.
XP 32-bit, Q6600, GTX 260 - 216 (driver 182.15)

I've been unable to successfully process any GPUGRID wu, but that's another story. I see the occasional aborted SETI CUDA wu abort, but I caught this one in the messages.

Any info.? Is this related to the VLAR issue?

Given your clues, I found the Task details 1320119915. It's not related to the VLAR issue, being a midrange "true angle range is : 0.432947". What happened was a failure to detect the GPU properly:

setiathome_CUDA: Found 1 CUDA device(s):
   Device 1 : Device Emulation (CPU) 
           totalGlobalMem = -1 

The emulation mode doesn't work, so eventually it gets to "Falling back to HOST CPU processing...", and that's a generic slow CPU version. The maximum elapsed time was set assuming the GPU would do the work, so wasn't enough for CPU fallback mode.

There are three other similar cases in that host's errored task list. I think others with similar problems have found updating to newer nVidia drivers helped.
                                                         Joe
ID: 922432 · Report as offensive
Profile Steve Dodd

Send message
Joined: 29 May 99
Posts: 23
Credit: 8,695,373
RAC: 1
United States
Message 922481 - Posted: 30 Jul 2009, 18:35:19 UTC - in response to Message 922432.  
Last modified: 30 Jul 2009, 18:36:27 UTC

Hi Josef,
I've been dealing with driver issues and GPUGRID. Have fallen back to this revision as it is the most stable at this time. I thought that 6.6.37 took care of this type of issue. I'll try a newer version of the driver and monitor. Thanks for the reply!
ID: 922481 · Report as offensive
Profile Sacaripasa
Avatar

Send message
Joined: 29 Dec 05
Posts: 13
Credit: 2,050,629
RAC: 1
United States
Message 922892 - Posted: 1 Aug 2009, 4:55:54 UTC

After posting this to another forum, which may have been wrong, I found this thread. Please help if you can....I am running the 190.&& nVidia drivers and just updated the CUDA .dll's to the neww ones posted via the lunatics web site. Got some fresh CUDA work and on my GeForce 9400 GT (driver version 19038, compute capability 1.1, 1024MB, est. 10GFLOPS) vid card, it is scheduled to take 13 HOURS to finish a batch! Any suggestions, or help IS GREATLY appreciated.
ID: 922892 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 922901 - Posted: 1 Aug 2009, 6:16:18 UTC - in response to Message 922892.  

After posting this to another forum, which may have been wrong, I found this thread. Please help if you can....I am running the 190.&& nVidia drivers and just updated the CUDA .dll's to the neww ones posted via the lunatics web site. Got some fresh CUDA work and on my GeForce 9400 GT (driver version 19038, compute capability 1.1, 1024MB, est. 10GFLOPS) vid card, it is scheduled to take 13 HOURS to finish a batch! Any suggestions, or help IS GREATLY appreciated.


Well 10Gflops is not fast for a graphics card.

Its only an estimated time. It will need to work on a couple of them to get a better idea of how long they take. If it isn't a VLAR it will probably be a lot better than the 13 hour estimate.
BOINC blog
ID: 922901 · Report as offensive
JohnDK Crowdfunding Project Donor*Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 28 May 00
Posts: 1222
Credit: 451,243,443
RAC: 1,127
Denmark
Message 923023 - Posted: 1 Aug 2009, 19:12:49 UTC - in response to Message 922892.  
Last modified: 1 Aug 2009, 19:13:34 UTC

After posting this to another forum, which may have been wrong, I found this thread. Please help if you can....I am running the 190.&& nVidia drivers and just updated the CUDA .dll's to the neww ones posted via the lunatics web site. Got some fresh CUDA work and on my GeForce 9400 GT (driver version 19038, compute capability 1.1, 1024MB, est. 10GFLOPS) vid card, it is scheduled to take 13 HOURS to finish a batch! Any suggestions, or help IS GREATLY appreciated.

You could add <flops> to app_info.xml as described here http://setiathome.berkeley.edu/forum_thread.php?id=53593
ID: 923023 · Report as offensive
FiveHamlet
Avatar

Send message
Joined: 5 Oct 99
Posts: 783
Credit: 32,638,578
RAC: 0
United Kingdom
Message 923032 - Posted: 1 Aug 2009, 20:00:07 UTC - in response to Message 923023.  

@ John
Is it possible to shutdown Boinc and add FLOPS lines in app_info.xml then restart Boinc,and not loose tasks.
I have about 1000 tasks to be processed and it will take a few days.Or
would it be better to run the tasks out and then alter the .xml?
I know the second option is the best way but I would like the timings a bit better.
ID: 923032 · Report as offensive
Fred W
Volunteer tester

Send message
Joined: 13 Jun 99
Posts: 2524
Credit: 11,954,210
RAC: 0
United Kingdom
Message 923038 - Posted: 1 Aug 2009, 20:58:24 UTC - in response to Message 923032.  

@ John
Is it possible to shutdown Boinc and add FLOPS lines in app_info.xml then restart Boinc,and not loose tasks.
I have about 1000 tasks to be processed and it will take a few days.Or
would it be better to run the tasks out and then alter the .xml?
I know the second option is the best way but I would like the timings a bit better.

Adding the flops line will not break anything as long as you are careful to re-save the .xml file as .xml and not as .txt.

F.
ID: 923038 · Report as offensive
FiveHamlet
Avatar

Send message
Joined: 5 Oct 99
Posts: 783
Credit: 32,638,578
RAC: 0
United Kingdom
Message 923048 - Posted: 1 Aug 2009, 21:54:16 UTC - in response to Message 923038.  

Thank's Fred
ID: 923048 · Report as offensive

Message boards : Number crunching : CUDA Taking Too Long To Process?


 
©2025 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.