Message boards :
Number crunching :
Only 1 GPU out of 4 is running !?!?
Message board moderation
Author | Message |
---|---|
DanHansen@Denmark Send message Joined: 14 Nov 12 Posts: 194 Credit: 5,881,465 RAC: 0 |
Hi, I've been building a Headless Linux CLI Multiple GPU System, based on Ubuntu 12.04 Server. Earlier on this year, I solved the problem using 12.10 Server/CUDA5.5, but now 12.10 isn't supported any more. So I had to find another combination. After some testing I found that the best solution was to use the old versions. Issues between e.g. Ubuntu 12.04.4+ and CUDA5.5+ forced med to use Ubuntu 12.04.3 and CUDA5.5. Never mind that now! My problem is that only 1 out of 4 GPU's run's. When using exactly the same set-up using 12.10 Server, it worked really good! But when using this 12.04.3 Server version and CUDA5.5 I had 2 issues. I'm not sure both of them are serious, but I'll explain it anyroad ;) After adding the CUDA 5.5 .deb package and installing it, I'm doing this to set-up minimum graphical environment/enable all GPU's. These are the errors I got: This is what I do: # apt-get install linux-image-extra-$(uname -r) x11-xserver-utils mesa-utils # modprobe nvidia # nvidia-xconfig --enable-all-gpus # cp /etc/X11/XF86Config /etc/X11/xorg.conf This is the errors: # apt-get install linux-image-extra-$(uname -r) x11-xserver-utils mesa-utils Reading package lists... Done Building dependency tree Reading state information... Done E: Unable to locate package linux-image-extra-3.8.0-29-generic E: Couldn't find any package by regex 'linux-image-extra-3.8.0-29-generic' And this one: # nvidia-xconfig --enable-all-gpus WARNING: Unable to locate/open X configuration file. WARNING: Unable to parse X.Org version string. Fact's: 1. It's a clean installation. Only vim-nox and lm-sensors has been installed. 2. I didn't upgrade the installation. If I had done that, then the 12.04.5 version would have messed things up, as you can see here. https://developer.nvidia.com/cuda-toolkit-55-archive The 12.04.4 update ruined the compatibility between 12.04 and CUDA5.5. I did try 12.04.5 using both CUDA5.5 and CUDA6.0 of course ;) If you have got an idea what might help, please let me know, so that I can get all 4 GPU's running again ;) .[/code] Project Headless CLI Linux Multiple GPU Boinc Servers Ubuntu Server 14.04.1 64bit Kernel 3.13.0-32-generic CPU's i5-4690K GPU's GT640/GTX750TI Nvidia v.340.29 BOINC v.7.2.42 |
DanHansen@Denmark Send message Joined: 14 Nov 12 Posts: 194 Credit: 5,881,465 RAC: 0 |
Hi, SOLVED THE PROBLEM!!! A conflict between the special set-up and the cc_config.xml was the problem! I did use cc_config.xml to allow remote control and multiple GPU's in this setup. And that coursed the problem. I don't know why , yet, but I hope I will find the answer. Just had to use the "old way" to allow remote control/GUI, by adding to "/etc/default/boinc-client" I thought I used cc_config in the other tests, and I did! But in those the value was set to: <use_all_gpus>0</use_all_gpus> which should had stopped the "extra" GPU's from running, or at least that's what I think ;) I think it has to do with this command in my ToDO: nvidia-xconfig --enable-all-gpus Anyway all 4 GPU's are running at 100% now ;) . Project Headless CLI Linux Multiple GPU Boinc Servers Ubuntu Server 14.04.1 64bit Kernel 3.13.0-32-generic CPU's i5-4690K GPU's GT640/GTX750TI Nvidia v.340.29 BOINC v.7.2.42 |
©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.