6.6.26 x32 CUDA issue

Author	Message
KW2E Send message Joined: 18 May 99 Posts: 346 Credit: 104,396,190 RAC: 34	Message 890404 - Posted: 2 May 2009, 2:33:52 UTC Last modified: 2 May 2009, 2:38:11 UTC I know that in newer versions of 6.6.x that there was a discussion about not using CUDA devices that were less capable than the primary CUDA device, but this is silly: 5/1/2009 8:26:00 PM CUDA device: GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) 5/1/2009 8:26:00 PM CUDA device (not used): GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) It's the exact same card! I cannot seem to find where I disable the "feature" of not using the less capable card. Any ideas? I'm going to revert to 6.6.23 for now. Rob ID: 890404 ·

KW2E Send message Joined: 18 May 99 Posts: 346 Credit: 104,396,190 RAC: 34	Message 890407 - Posted: 2 May 2009, 2:47:44 UTC - in response to Message 890406. Careful Vic, it's just alpha software. There are bound to be issues. I know this so it's no big deal to me. I just revert back to .23 and call it good for now. It does seem rather silly though. :) Just reporting an issue. Rob ID: 890407 ·

KW2E Send message Joined: 18 May 99 Posts: 346 Credit: 104,396,190 RAC: 34	Message 890422 - Posted: 2 May 2009, 4:14:56 UTC - in response to Message 890421. The AMD? Yes and no. I had it running for a bit but I have since set it back on the shelf. Plenty of heat generators in this office as it is. :) Thanks Vic, Rob ID: 890422 ·

Paul D. Buck Volunteer tester Send message Joined: 19 Jul 00 Posts: 3898 Credit: 1,158,042 RAC: 0	Message 890480 - Posted: 2 May 2009, 9:25:08 UTC Known issue because of a new feature added to prevent less capable CUDA cards from being used. 6.6.25 has a new cc_config flag to use all GPUs in options. UCB is aware and the issue is that the second card has a 64K block of memory less than the other cards. Why, not known, driver issue? Video memory reserve? Gremlins? ID: 890480 ·

MarkJ Volunteer tester Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5	Message 890505 - Posted: 2 May 2009, 11:44:26 UTC - in response to Message 890404. I know that in newer versions of 6.6.x that there was a discussion about not using CUDA devices that were less capable than the primary CUDA device, but this is silly: 5/1/2009 8:26:00 PM CUDA device: GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) 5/1/2009 8:26:00 PM CUDA device (not used): GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) It's the exact same card! I cannot seem to find where I disable the "feature" of not using the less capable card. Any ideas? I'm going to revert to 6.6.23 for now. Rob I got exactly the same thing. I used the cc_config flag <use_all_gpus> to get them going. As Paul has said apparently its reporting a different memory size. Given the cards have the same memory it could be the drivers. I had 182.50 drivers installed. It could be a bug in the detection logic since 6.6.23. Trak ticket 888 has been raised for this and the email list notified. BOINC blog ID: 890505 ·

MarkJ Volunteer tester Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5	Message 890711 - Posted: 3 May 2009, 8:21:52 UTC - in response to Message 890505. I know that in newer versions of 6.6.x that there was a discussion about not using CUDA devices that were less capable than the primary CUDA device, but this is silly: 5/1/2009 8:26:00 PM CUDA device: GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) 5/1/2009 8:26:00 PM CUDA device (not used): GeForce GTX 260 (driver version 18122, compute capability 1.3, 896MB, est. 117GFLOPS) It's the exact same card! I cannot seem to find where I disable the "feature" of not using the less capable card. Any ideas? I'm going to revert to 6.6.23 for now. Rob I got exactly the same thing. I used the cc_config flag <use_all_gpus> to get them going. As Paul has said apparently its reporting a different memory size. Given the cards have the same memory it could be the drivers. I had 182.50 drivers installed. I have confirmed the memory size reporting on GPUgrid (it reports the memory for each cuda device in the work units). Its reporting 64k less memory on the 1st device. Beta drivers 185.81 don't show the missing 64k so I have been told. I haven't tried them so can't verify this or not. They also give cuda version 2.2 with a slight reduction in run times and increased heat produced by the cards, so I have read. BOINC blog ID: 890711 ·

Valerie Chilton Volunteer tester Send message Joined: 14 Jan 08 Posts: 53 Credit: 20,682,710 RAC: 0	Message 890889 - Posted: 3 May 2009, 17:52:54 UTC I'm having much the same thing with the 64 bit. One of the cars isn't used. What file and what changes do I need to make to enable the 2nd card ? And after the spanking comes... Ni ! ID: 890889 ·

Valerie Chilton Volunteer tester Send message Joined: 14 Jan 08 Posts: 53 Credit: 20,682,710 RAC: 0	Message 890900 - Posted: 3 May 2009, 18:35:17 UTC Changing back to the older ver fixed the prob. And after the spanking comes... Ni ! ID: 890900 ·

MarkJ Volunteer tester Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5	Message 891145 - Posted: 4 May 2009, 13:00:40 UTC - in response to Message 890889. I'm having much the same thing with the 64 bit. One of the cars isn't used. What file and what changes do I need to make to enable the 2nd card ? in cc_config.xml you'll need: <options> <use_all_gpus>1</use_all_gpus> </options> Oh and I tried the 185.81 drivers. I had GPUgrid work crapping out all over the place. Uninstalled it. Reinstalled 182.50 drivers, trashed all the cuda work it had. Now crunching on some fresh work. BOINC blog ID: 891145 ·

©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.