Message boards :
Number crunching :
Ubuntu 11.10 not using both NVIDIA GPU's
Message board moderation
Author | Message |
---|---|
Alaun Send message Joined: 29 Nov 05 Posts: 18 Credit: 9,310,773 RAC: 0 |
I have two GTX 460's. BOINC recognizes both of them and seems like it's trying to use both of them. Running (0.25 CPUs + 0.50 NVIDIA GPUs (device 0)) Running (0.25 CPUs + 0.50 NVIDIA GPUs (device 0)) Running (0.25 CPUs + 0.50 NVIDIA GPUs (device 1)) Running (0.25 CPUs + 0.50 NVIDIA GPUs (device 1)) Seems good but when I look at the load/temps, one card is maxed out and other is at idle. What I have tried: -Dummy plug. It shows up in X Server. It's enabled as 'Seperate X Screen' -use_all_gpus. It shows up in the event log, but makes no difference. -Flipping the cards around and all combinations of connectivity. -Enabling 'Xinerama'. This causes mayhem and doesnt work. -deleting the contents of the xorg.conf file (to get rid of xinerama) Any ideas?? System: Intel Quad 2.4GHZ Two GTX 460's Ubuntu 11.10 (64 bit) BOINC 6.12.33 Seti CUDA app from Crunchers Anonymous setiathome-6.11.x86_64-pc-linux-gnu__cuda32 Current NVIDIA drivers and CUDA toolkits. |
Khangollo Send message Joined: 1 Aug 00 Posts: 245 Credit: 36,410,524 RAC: 0 |
There ia a known issue with that old GPU application you are using (it doesn't support multiple GPUs). Try the new x41g application from Lunatics. It is much faster. You don't need those dummy plugs, extending desktops or similar dirty workarounds wintendo users have to endure. X server is also not needed (as long as you force kernel driver to stay loaded with nvidia-smi -pm 1 ). |
Alaun Send message Joined: 29 Nov 05 Posts: 18 Credit: 9,310,773 RAC: 0 |
It works! Thank you sir~ Both cards are loaded up and sounding great. |
Terror Australis Send message Joined: 14 Feb 04 Posts: 1817 Credit: 262,693,308 RAC: 44 |
..... X server is also not needed (as long as you force kernel driver to stay loaded with nvidia-smi -pm 1 ). Thanks Khangollo, this could also be the answer to a problem I've been having. A couple of questions. Does this still give you a desktop or is it console operation only without the X server running ? How do I add that command to the start up ? and, do I need to delete the xorg.conf file as well ? T.A. |
Khangollo Send message Joined: 1 Aug 00 Posts: 245 Credit: 36,410,524 RAC: 0 |
Does this still give you a desktop or is it console operation only without the X server running ? This command doesn't actually start the X server or graphics. It just loads the nvidia kernel module (and "stays" in console) - so that BOINC can detect and use the GPU. When you start X server it gets loaded automatically. But with no X server, you need to load it with this command. It also prevents the driver from being unloaded when it is no longer in use for a while. If you do not wish to boot directly into X, you need to edit the file /etc/inittab and change id:5:initdefault:to id:3:initdefault: Of course you can then still startx from console and that won't disrupt CUDA. How do I add that command to the start up ? ...depends on the distribution and where the boinc is started on boot. nvidia-smi -pm 1 needs to be called before boinc is started. In Fedora I start boinc in script /etc/rc.d/init.d/boinc (in Debian-based distros it should be /etc/init.d/boinc ) Edit that file, just after block of # comments at the beginning of file, insert a new line: /usr/bin/nvidia-smi -pm 1 > /dev/null 2>&1 and, do I need to delete the xorg.conf file as well ? No. |
©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.