|
1)
Message boards :
BOINC client :
NO CUDA DEVICE FOUND
(Message 24220)
Posted 11 Apr 2009 by David Post: Make sure the /etc/security/console.perms.d/* file that sets the /dev/nvidia* devices to root.video is setting the mode as 0660 in both cases. If it is, I'd also check the group and mode set in /etc/udev/rules.d/* for the nvidia devices. |
|
2)
Message boards :
BOINC client :
NO CUDA DEVICE FOUND
(Message 24034)
Posted 31 Mar 2009 by David Post: I've just solved my CUDA problem on Fedora 10. This may also work for Gentoo. There are two problems, firstly the RPM doesn't include the libcudart.so in /var/lib/boinc which is used for the data store for BOINC. I downloaded the latest CUDA SDK from nVidia and put a link in /var/lib/boinc to the libcudart.so in the SDK. The second problem is that the boot script that starts the BOINC client runs the process as the "boinc" user. This user does not have permission to access the /dev/nvidia* devices and so cannot find the CUDA devices. To fix it: groupadd -r gpu usermod -G gpu boinc Then create a file with the following content as /etc/security/console.perms.d/60-nvidia-gpu.perms <gpu>=/dev/nvidia* <console> 0660 <gpu> 0660 root.gpu You can then logout and login again, or pam_console_apply -r /dev/nvidia*; pam_console_apply /dev/nvidia* to reset the device permissions. Restart boinc client, e.g. service boinc-client restartor /etc/init.d/boinc stop; /etc/init.d/boinc start Then check the messages to see if BOINC can now see your CUDA devices. |
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.