Message boards :
Number crunching :
Not all cuda devices used
Message board moderation
Author | Message |
---|---|
Valerie Chilton Send message Joined: 14 Jan 08 Posts: 53 Credit: 20,682,710 RAC: 0 |
Greetings! I'm running Boinc 6.6.26 and have a Nvidia 260 and a have added a Nvidia 250 int this (4720228) machine. When I start up boinc it now sees both cards as a 250 and says one of them is not used. The GFlops for the 260 was much higher than the 250 but both are now rated at 84GFLOPS. Any ideas how to enable the new device and how to make it see the 260 with its' higher gflops as the right card ? And after the spanking comes... Ni ! |
Raistmer Send message Joined: 16 Jun 01 Posts: 6325 Credit: 106,370,077 RAC: 121 |
Try to use older BOINC like 6.6.20. New ones seems have very doubtful "feature" to enable only best card on host.... I consider it as direct act of sabotage ;) |
Valerie Chilton Send message Joined: 14 Jan 08 Posts: 53 Credit: 20,682,710 RAC: 0 |
Changing back did fix the problem, I agree, the new "feature" really bites. And after the spanking comes... Ni ! |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
It is known design feature. Idea is sound in theory as it will prevent using lesser card so video redraw not effected like when pure opt out. To use all GPU's add the line <use_all_GPUs> to cc_config. Here is a template from Raistmer. <cc_config> <options> <ncpus>NUMBER_OF_LOGICAL_PROCESSORS+NUMBER_OF_GPUs</ncpus> <use_all_gpus>1<use_all_gpus> </options> </cc_config> "The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov |
zoom3+1=4 Send message Joined: 30 Nov 03 Posts: 65775 Credit: 55,293,173 RAC: 49 |
It is known design feature. Idea is sound in theory as it will prevent using lesser card so video redraw not effected like when pure opt out. For 6.6.20 and newer use this one instead, As <npus>NUMBER_OF_LOGICAL_PROCESSORS+NUMBER_OF_GPUs</ncpus> is not needed with 6.6.20 or Newer. <cc_config> <options> <use_all_gpus>1<use_all_gpus> </options> </cc_config> The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
Sorry, thanks! Using teamwork mod here, as it's works the best for me. Forgot all about using stock cuda scheduling. "The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov |
Westsail and *Pyxey* Send message Joined: 26 Jul 99 Posts: 338 Credit: 20,544,999 RAC: 0 |
DP "The most exciting phrase to hear in science, the one that heralds new discoveries, is not Eureka! (I found it!) but rather, 'hmm... that's funny...'" -- Isaac Asimov |
Rob.B Send message Joined: 23 Jul 99 Posts: 157 Credit: 1,439,682 RAC: 0 |
My recent hellish move to 6.6.20 had similar issues, with Boinc only utilizing the fater card. I deinstalled the slower card, upgraded the NVIDIA drivers to the latest version (18250) then reinstalled the slower card. The NVIDIA drivers replaced were quite upto date being only about four or five weeks old. 6.6.20 is now happily using both cards with the new driver. [See snipped from the log file.] 04/05/2009 07:39:31 CUDA devices: GeForce 9800 GTX/9800 GTX+ (driver version 18250, CUDA version 1.1, 512MB, est. 86GFLOPS), GeForce 9400 GT (driver version 18250, CUDA version 1.1, 512MB, est. 8GFLOPS) Rob |
Mudbankkeith Send message Joined: 26 May 07 Posts: 8 Credit: 5,911,186 RAC: 0 |
Same problem here. rollback to version 6.6.20 fixed the problem. |
Mark Henderson Send message Joined: 9 Mar 02 Posts: 41 Credit: 3,964,939 RAC: 0 |
Had the same problem with "not used" on my 2nd card with 6.6.28. Had been using 6.6.20 but was reading on the GPU Grid boards that 6.6.23 was better. Im now using 6.6.23 with no problems reading both cards. |
MarkJ Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5 |
It was introduced in 6.6.24. After complaining Dr A revised it to allow a 30% variance in memory from the 1st device to any others. If they are totally different then you need to either drop back to 6.6.23 or use the cc_config flag. Personally i'd suggest 6.6.23 over 6.6.20 as it has a couple of cuda fixes. It depends on if you want to run the official version or the development version. BOINC blog |
©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.