CUDA working on ubuntu-desktop not on ubuntu-server

I run boinc, gpugrid with my graphics card. I have nVidia 9800 GTX+. First I run the boinc with cuda on ubuntu desktop edition, then I installed the ubuntu server on the same computer (without gui) and cuda doesn’t work.

Here is how I installed it on Ubuntu desktop:

  1. downloaded the drivers: NVIDIA-Linux-x86_64-190.32-pkg2.run

  2. killed gdm, and installed them by runing

sudo ./NVIDIA-Linux-x86_64-190.32-pkg2.run
  1. after restart cuda worked, boinc worked

On ubuntu server I did the same thing without killing gdm, since there is no gui. Drivers installation worked fine, but boinc could not find cuda.

I also checked with some program named deviceQueryDrv.

While on desktop I got the info about the drivers and cuda, on server I got the error:

CUDA Device Query (Driver API) statically linked version 

Cuda driver error 3 in file 'deviceQueryDrv.cpp' in line 76.

Any suggestions?

If you don’t run X, then you need to run a script at startup to create the /dev/nvidia* device files required for CUDA:

http://forums.nvidia.com/lofiversion/index.php?t52629.html

My issue may be related.

I have a motherboard with onboard graphics and an NVIDIA card in the PCIEx16 slot.

I have a fresh install of Ubuntu 9.04

I want to load the NVIDIA driver with my monitor plugged into the on-board graphics to run X.

So far I can only get CUDA to work when I plug my monitor into the GPU.

I tried running the shell script but get this error:

michael@GPUbuntu:~/sh$ sudo ./startCuda.sh

./startCuda.sh: line 8: /sbin/lspci: No such file or directory

./startCuda.sh: line 9: /sbin/lspci: No such file or directory

mknod: `/dev/nvidiactl': File exists

I got it to work.

Just a slight modification to the shell script

in Ununtu Desktop 9.04, lspci is located in /usr/bin/ not /sbin/

here is the code:

#!/bin/bash

modprobe nvidia

if [ "$?" -eq 0 ]; then

# Count the number of NVIDIA controllers found.

N3D=`/usr/bin/lspci | grep -i NVIDIA | grep "3D controller" | wc -l`

NVGA=`/usr/bin/lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`

N=`expr $N3D + $NVGA - 1`

for i in `seq 0 $N`; do

mknod -m 666 /dev/nvidia$i c 195 $i;

done

mknod -m 666 /dev/nvidiactl c 195 255

else

exit 1

fi

Hi Guys,

I’ve been fighting this problem for more than a month. We are trying to install a Headless Ubuntu Server 12.04 Boinc GPU Cruncher. I tried everything. Many different ways to discard the nouveau driver and then install the Nvidia driver, I tried the solution where the driver modifies things for you. The only way we succeeded was to install the Desktop Edition.
I heard from a lot of places, that it’s not possible to crunch data using the GPU on Ubuntu Server 12.04, because there’s no GUI or X-server.
But I think you just might solved the case! Didn’t you? Would you please show us how to do this? we are pretty many guys from Berkely.edu who have strugled with this for some time now.

Here’s the hardware:
Intel i5-3470K
Asus P8H77-M Pro
8Gb Kingston HyperX Genesis X2 Grey S.
Asus GT640-1GD5-L PCIe 2.0 x16
Asus GT640-1GD3-L PCIe 3.0 x16 (this will not be installed before we succeed using 1 GPU, if we succeed)

Clean installation of:
Ubuntu Server 12.04.4 64bit

All CLI, Rack mounted servers. No GUI at all. Uses SSH to remotely control the Boinc Servers ;)

I’m really exited now, I think you now how to do this, if I’m not mistaking. You can see for how long time we tried here:

http://setiathome.berkeley.edu/forum_thread.php?id=73032&postid=1484254#1484254
http://ubuntuforums.org/showthread.php?t=2204590&page=2&p=12924359
http://setiathome.berkeley.edu/forum_thread.php?id=74153&postid=1484338#1484338
http://www.overclock.net/t/1467918/ubuntu-server-12-04-4-64bit-boinc-using-gpu-from-geforce-gt610-to-crunch-data#post_21889886
http://www.hardwarecanucks.com/forum/hardwarecanucks-f-h-team/58582-experiment-whats-minimal-install-get-nvidia-boinc-crunching.html#post757903

Please help me solve this matter. I can try the things you did, but there’s a couple of things I would like to ask you.

We are looking much forward to hear from you ;)

Kind Regards,
Dan Hansen

Last week I installed Ubuntu 12.04 on a new machine with a GTX Titan using debootstrap, which should be even more minimal than Ubuntu Server. The only extra’s I installed was the Linux kernel headers and gcc. After that I used the Nvidia drivers and toolkit from the package manager. Maybe you are only missing the kernel headers? Below are the commands I used, hopefully they will work for you too:

sudo apt-get install build-essential linux-headers-`uname -r`
wget http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1204/x86_64/cuda-repo-ubuntu1204_5.5-0_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1204_5.5-0_amd64.deb
sudo apt-get update
sudo apt-get install cuda

Line #1 installs the dependencies for the driver and CUDA toolkit. “build-essential” contains things like gcc, make, etc. and “linux-headers-uname -r” installs the header files for the Linux kernel you are currently running. Line #2 downloads the “cuda-repo-ubuntu1204_5.5-0_amd64.deb” file from the Nvidia website, while line #3 adds this file to the Ubuntu package manager. Line #4 updates the list of packages in the package manager, such that it now includes Nvidia’s CUDA packages. Line #5 installs the cuda package, which contains the driver and the toolkit. It will also install any dependencies (except for the kernel headers), so this may take some time.

Hi Gert-Jan,

Thank you for writing me ;) Thanks my friend. You have no idea how long I have searched for some help on this matter ;)
I almost can’t wait to try those things from your post! I would very much like to tell you about my attempts too. But I have just one more class at Uni before I’m done for the day ;)

Just a short reply for now ;)
In the last attempt I installed “gcc” too, and then blacklisted nouveau and some others as well. Installed the Nvidia driver and let it modify “something” can’t remember. The GPU was accepted but didn’t do any work.

My question:

  1. Do you think it’s possible to use your ToDo on a Ubuntu Server 12.04 64bit Installation?
  2. What’s you number 3? Install “cuda-repo-ubuntu1204_5.5-0_amd64.deb” ?? If it is, can you show me how a .deb is installed? I usually use .run / .sh :o
  3. When you in “6” say “install cuda” is it this one then? Link: https://developer.nvidia.com/cuda-downloads Ubuntu 12.04 .run

Sorry for the newbie questions :o - I’ve only worked with Ubuntu for little more than a year ;)

I’m very thankful Gert-Jan !!! Is it all right if I tell the guys at Berkeley.edu about you and what you showed me? I would really enjoy this very much ;)

Kind Regards,
Dan Hansen

Hi Dan, you’re welcome! Feel free to share this information, it is out here on the forum to share ;) Line #3 was empty, I removed it (don’t know how it got there in the first place). I have also added some more descriptions to my original post. I’m pretty sure this will also work for your Ubuntu Server 12.04 64bit installation.

I got most of the information from the NVIDIA CUDA Getting Started Guide for Linux and I only used the .deb file, no .sh or .run files. The reason to go for the .deb files instead of the traditional .run files is that the .deb files can track dependencies to other files / libraries and install them as required, which makes installing CUDA on Linux a lot easier. When installing via the .deb files, it is not necessary to blacklist the nouveau drivers manually, the installation process will do this for you.

After installing I modified the PATH and LD_LIBRARY_PATH environment variables for all users by modifying the “/etc/environment” file:

PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/cuda-5.5/bin"
LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib

Hi Gert-Jan,

I’m so happy. I’m so so happy. This is the first serious suggestion I found. Can’t wait ;)

When installing the .deb file, is it then updated when using the apt-get update command like all other packages?

Do you use the Boinc repository or do you install it with the ubuntu package “apt-get install boinc-client”??? I’m still not sure about the difference. I did read a lot about repository and Ubuntu, but it didn’t get through my thick scull ;)

“…When installing via the .deb files, it is not necessary to blacklist the nouveau drivers manually, the installation process will do this for you…”
I’m so f… happy to hear this, you have no idea!! I’ve been working through the night, one night after another ;(

PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/cuda-5.5/bin"
LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib

Will I need to do this as well? Does the Server Edition has this path/directories??
I read this on the documentation site:
“…The LD_LIBRARY_PATH variable needs to contain /usr/local/cuda-5.5/lib on a 32-bit system, and /usr/local/cuda-5.5/lib64 on a 64-bit system…”

apt-get install build-essential linux-headers-`uname -r`

This I don’t use when I have the Ubuntu Server Edition 12.04.4 64bit installed, right?

“…http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1204/x86_64/cuda-repo-ubuntu1204_5.5-0_amd64.deb…”
When installing this, I don’t need the normal Nvidia driver as well, right? This IS the driver needed?? Or the CUDA-toolkit, right??

I want to be sure, before I get on with it ;) Don’t want to f… it up ;) Man, have I waited for this!!!

Where are you from? Gert-Jan sounds like we are neighbours. I’m from Denmark ;)

Kind Regards,
Dan Hansen

I don’t use boinc, so can’t help you with that

  1. After installing the driver and toolkit, it is best to update the PATH and LD_LIBRARY_PATH such that the nvcc compiler and CUDA libraries can be found by the operating system.

  2. Not sure if you need it with Ubuntu server, or if it is already installed. I would do it anyway, it can’t hurt, it just installs some header files.

  3. Right, this is an all-in-one package, including the driver and toolkit. No need for any other driver.

The Ubuntu package repository is just a big app-store. By adding the .deb file (command #3) you add cuda-5.5 to the store, after which you can install it. apt-get update / upgrade should include updates to CUDA, if Nvidia releases any (I don’t think it will automatically upgrade from 5.5. to 6.0 though, you would need to download another .deb file).

We’re not neighbours, but we’re close, I’m from the Netherlands ;)

Hi Gert-Jan,

Thanks! I’m doing it right now ;)

Regarding the PATH stuff. Is your example with or without commands? It’s because I read something about changing paths from “http://docs.nvidia.com/cuda/cuda-getting-started-guide-for-linux/index.html#package-manager-installation” and they are using the “export command”!? Or do I just do it as you wrote and edit the file /etc/environment (I use VI):

PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/cuda-5.5/bin"
    LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib

Another thing is, they suggest that there’s a difference between 32bit and 64bit systems:
“…The LD_LIBRARY_PATH variable needs to contain /usr/local/cuda-5.5/lib on a 32-bit system, and /usr/local/cuda-5.5/lib64 on a 64-bit system…”
Then I have to use the 64bit method, right? So I “export” or “VI” (edit) my /etc/environment file!?


05:11
OK, I just edited the file /etc/environment and added to the path: “/usr/local/cuda-5.5/bin”. And then I added in the next line: “LD_LIBRARY_PATH=/usr/local/cuda-5.5/lib64”. If it doesn’t work I can just edit it again and remove “64” from line 2 ;)

Well, regarding the “export command” I guess it does the same as when we edit it directly. I gets me thinking of the command: “echo “FQDN” > /etc/hostname”. The export command does the same, add’s to the file I guess!?!?

Well, this is very nerve wrecking ;) Let’s test it and see if it works. ;)
I’ll be back tomorrow ;)


05:39
Well, it is kind of tomorrow now ;) It didn’t work, the GPU didn’t get recognized by Boinc. I’m so sad :( It sounded so g… d… good!! “nouveau” being “stopped” automatically and stuff. It really looked so good. It’s a clean installation of Ubuntu Server 12.04. Didn’t do anything other than modify it to a static IP. No driver installation or anything. It was a clean installation.
A error command during boot-up: “Starting … load fallback graphics devices … [fail]” or something like that.


05:59
OK!! Look here. Maybe there’s still hope. From the Nvidia CUDA site it says:
https://developer.nvidia.com/cuda-downloads
*** The CUDA 5.5 Debian packages are not compatible with Ubuntu 12.04 after the 12.04.4 LTS update. Please use the .run installer instead.

But!! If we have to use the .run file, then what about all the great advantages of the .deb method? And I’m so DAMN disappointed in Ubuntu, or at least I’m getting there!! Can you help me Gert-Jan? Please be specific if you can help me. What do we do? How does it affect the .deb way to do it? Can we run the .run file and then that’s it? Or do I have to redo something? Update? Reinstall?

CUDA Information also says:

2.5.2. Runfile Installation
If you are using Grub2 as a bootloader, you may also need to edit its config to prevent Nouveau from loading. Add “rdblacklist=nouveau nouveau.modeset=0” to the end of the GRUB_CMDLINE_LINUX entry in /etc/default/grub. Then, remake your Grub configuration by running:

# grub2-mkconfig -o /boot/grub2/grub.cfg

:(


06:22

Hi Gert-Jan… It’s not looking good :(

I tried this:

wget http://developer.download.nvidia.com/compute/cuda/5_5/rel/installers/cuda_5.5.22_linux_64.run
sh cuda_5.5.22_linux_64.run

This is the result:

===========
= Summary =

Driver: Installation Failed
Toolkit: Installation skipped
Samples: Installation skipped

Logfile is /tmp/cuda_install_1689.log
root@myprettyfuckedupcommodore64:/home/myusername#

Kind Regards,
Dan

Looks like for some reason the driver could not be installed. The reason is probably hidden somewhere in the log file (e.g. /tmp/cuda_install_1689.log). Most likely you are missing some files (kernel header files?), which files exactly you can find in the log file. Can you upload the logfile, so we can have a look?

I’m using the .deb file and I installed Ubuntu 12.04.4, so although it shouldn’t work according to the Nvidia website, it did work for me.

The last time I installed a using a .run file was over a year ago, with Ubuntu 12.04.1 and CUDA 5.0. I wrote down the things I did back than over here. When you follow the steps in the documentation, where does it go wrong? The first thing you want to get working is the installer (.run) to say Driver, Toolkit & Samples installation successful. After that you can look at issues like grub and PATH, if they are a problem.

Hi Gert-Jan,

I viewed the log file, " vi /tmp/cuda_install_1689.log" and it was empty ???

I did try it just like you wrote, using the .deb file. And everything went well, no errors or warnings during the setup.

And it worked for you, 64bit and v.12.04.4 ??

There’s 3 other files in /tmp ! I did try 3 more times. Here’s the content of the first one:

/tmp# vi cuda_install_1704.log
Using more to view the EULA.
Installing the NVIDIA display driver...
Verifying archive integrity... OK

Uncompressing NVIDIA Accelerated Graphics Driver for Linux-x86_64 319.37......................................................................................................................................................................

WARNING: One or more modprobe configuration files to disable Nouveau are

         already present at:

         /etc/modprobe.d/nvidia-installer-disable-nouveau.conf.  Please be

         sure you have rebooted your system since these files were written.

         If you have rebooted, then Nouveau may be enabled for other

         reasons, such as being included in the system initial ramdisk or

         in your X configuration file.  Please consult the NVIDIA driver

         README and your Linux distribution's documentation for details on

         how to correctly disable the Nouveau kernel driver.

Welcome to the NVIDIA Software Installer for Unix/Linux

License accepted by command line option.

Installing NVIDIA driver version 319.37.

For some distributions, Nouveau can be disabled by adding a file in the

modprobe configuration directory.  Would you like nvidia-installer to

attempt to create this modprobe file for you? (Answer: Yes)

One or more modprobe configuration files to disable Nouveau have been

written.  For some distributions, this may be sufficient to disable

Nouveau; other distributions may require modification of the initial ramdisk.  Please reboot your 

system and attempt NVIDIA driver installation

again.  Note if you later wish to reenable Nouveau, you will need to delete

these files: /etc/modprobe.d/nvidia-installer-disable-nouveau.conf

Would you like to register the kernel module sources with DKMS? This will

allow DKMS to automatically build a new module, if you install a different

kernel later. (Answer: No)

Performing CC sanity check with CC="cc".

Kernel source path: '/lib/modules/3.11.0-15-generic/build'

Kernel output path: '/lib/modules/3.11.0-15-generic/build'

Performing rivafb check.

Performing nvidiafb check.

Performing Xen check.

Cleaning kernel module build directory.

Building kernel module:

"cuda_install_1704.log" 115L, 16123C

Did this “# /etc/modprobe.d/nvidia-installer-disable-nouveau.conf”

Result: “permission denied”

Thanks for helping ;)

Kind Regards,
Dan

According to these messages, you should run the installer, as you already did, reboot, and then run the installer again. The first time you run the installer it tries to disable nouveau by writing a file called “/etc/modprobe.d/nvidia-installer-disable-nouveau.conf”. After rebooting nouveau should no longer be running, and when you run the installer again it should continue. If nouveau is still running after the reboot you have to stop it somehow, maybe the “NVIDIA driver README and your Linux distribution’s documentation” know what to do then. To see if nouveau is still running, you can run lsmod to see which modules are currently loaded.

The installer generates a log file with a different name every time you run it, so you have to check which is the last one.

Hi,

I blacklistet nouveau in “/etc/modprobe.d/blacklist.conf” and ran the installation file again.

Same error. New log file:

Using more to view the EULA.
Installing the NVIDIA display driver...
Verifying archive integrity... OK

Uncompressing NVIDIA Accelerated Graphics Driver for Linux-x86_64 319.37......................................................................................................................................................................

WARNING: One or more modprobe configuration files to disable Nouveau are

         already present at:

         /etc/modprobe.d/nvidia-installer-disable-nouveau.conf.  Please be

         sure you have rebooted your system since these files were written.

         If you have rebooted, then Nouveau may be enabled for other

         reasons, such as being included in the system initial ramdisk or

         in your X configuration file.  Please consult the NVIDIA driver

         README and your Linux distribution's documentation for details on

         how to correctly disable the Nouveau kernel driver.

ERROR: Unable to build the NVIDIA kernel module.

Welcome to the NVIDIA Software Installer for Unix/Linux

License accepted by command line option.

Installing NVIDIA driver version 319.37.

For some distributions, Nouveau can be disabled by adding a file in the

modprobe configuration directory.  Would you like nvidia-installer to

attempt to create this modprobe file for you? (Answer: Yes)

One or more modprobe configuration files to disable Nouveau have been

written.  For some distributions, this may be sufficient to disable

Nouveau; other distributions may require modification of the initial

ramdisk.  Please reboot your system and attempt NVIDIA driver installation

again.  Note if you later wish to reenable Nouveau, you will need to delete

these files: /etc/modprobe.d/nvidia-installer-disable-nouveau.conf

Would you like to register the kernel module sources with DKMS? This will

allow DKMS to automatically build a new module, if you install a different

kernel later. (Answer: No)

Performing CC sanity check with CC="cc".

Kernel source path: '/lib/modules/3.11.0-15-generic/build'

Kernel output path: '/lib/modules/3.11.0-15-generic/build'

Performing rivafb check.

Performing nvidiafb check.

Performing Xen check.

Cleaning kernel module build directory.

Building kernel module:

ERROR: Installation has failed.  Please see the file

       '/var/log/nvidia-installer.log' for details.  You may find

       suggestions on fixing installation problems in the README available

       on the Linux driver download page at www.nvidia.com.

The driver installation is unable to locate the kernel source. Please make sure that the kernel source packages are installed and set up correctly.
If you know that the kernel source packages are installed and set up correctly, you may pass the location of the kernel source with the '--kernel-source-path' flag.

===========
= Summary =
===========

Driver:   Installation Failed
Toolkit:  Installation skipped
Samples:  Installation skipped

Logfile is /tmp/cuda_install_1710.log

It’s a little tricky to VI this log file. You have to scroll down using the cursor, and all sorts of weird stuff is in the middle of the file.

But, Gert-Jan. I think we will have to be smart now. This is really a time kiler, and I was thinking. Why not do what you did?
So, which version of ubuntu did you use? “debootstrap” is what? Did you make a USB disk or a CD/DVD and which version of ubuntu? Desktop Installation og Server Installation or something completely different ?

I’ll try to do what you did! We can agree on this, right?

Ubuntu non-graphical, no GUI
64bit OS

Then what? ;) Show me the money Mr. Gert-Jan ;)

You can also use nano to inspect log files, it is a little easier than vi. Also check this log file: /var/log/nvidia-installer.log

I checked the steps I did to install Ubuntu using debootstrap, but it is pretty customized. A more general description can be found here. I would recommend not using this, as it is much more complicated than installing Ubuntu (server) the regular way.

Right now I would first check the /var/log/nvidia-installer.log file. If the errors there can’t be fixed and it still does not work I would try a desktop Ubuntu (64 bit), this should work easily as everyone is using it. If this works, you can try again with Ubuntu server, as it should be very similar, maybe it is missing a couple of things which you can then install manually.

Hi Gert-Jan,

“…The last time I installed a using a .run file was over a year ago, with Ubuntu 12.04.1 and CUDA 5.0. I wrote down the things I did back than over here…”

I looked at it, its a GUI environment. I’m all CLI ;)

“…When you follow the steps in the documentation, where does it go wrong?..”

I’m not sure which of the commands to use. I’m not very good at Linux, not yet at least. 1 year doing it so far ;)
Output from the guide:
“…Consult your distribution’s documentation to find out how to properly exit the GUI. …”
“…Exit the GUI if you are in a GUI environment by pressing Ctrl-Alt-Backspace…”
“…Restart the GUI environment (using the command startx or init 5 or sudo /etc/init.d/gdm start or the equivalent command on your system)…”
etc. etc.
I’m running a non-graphical environment. No GUI! This is apparently hard for the developers of ubuntu to understand. I looked at so many sites and in so many books for the last 1 1/2 month and its always the same. Every system is based on GUI.
I just can’t understand why its so unusual, wanting to build a number cruncher with less rather than more!?!?

I’ tried another thing before you wrote me. I tried this:

Clean installation of Ubuntu Server 12.04.4, and then

# apt-get update
# apt-get upgrade
# apt-get install gcc
# apt-get install make
# wget http://us.download.nvidia.com/XFree86/Linux-x86_64/331.49/NVIDIA-Linux-x86_64-331.49.run
# chmod 0755 NVIDIA-Linux-x86_64-331.49.run
# ./ NVIDIA-Linux-x86_64-331.49.run
# vi /etc/init.d/nvidia_cuda

inserted this:

#!/bin/bash

PATH=/sbin:/bin:/usr/bin:$PATH

/sbin/modprobe nvidia

if [ "$?" -eq 0 ]; then

   # Count the number of NVIDIA controllers found.
   N3D=`/usr/bin/lspci | grep -i NVIDIA | grep "3D controller" | wc -l`
   NVGA=`/usr/bin/lspci | grep -i NVIDIA | grep "VGA compatible controller" | wc -l`

   N=`expr $N3D + $NVGA - 1`
   for i in `seq 0 $N`; do
      /bin/mknod -m 666 /dev/nvidia$i c 195 $i;
   done

   /bin/mknod -m 666 /dev/nvidiactl c 195 255

else
   exit 1
fi

# chmod 0755 /etc/init.d/nvidia_cuda
# update-rc.d nvidia_cuda defaults
# reboot
# apt-get install boinc-client

attached to projects etc.

And, as a matter of fact the GPU now appears in Boinc. I checked the heat of the GPU, because I’ve installed the Nvidia driver earlier on in other tests. And back then it ran for 10min. and the stopped. Here, the same thing happens. Look at this:

The command every minute or so. And as you can see, it runs for a time, then it suddely stops and the temp. drops!

# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 37 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 37 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 38 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 38 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 39 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 39 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 40 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 40 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 40 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 40 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 43 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 43 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 37 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 36 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 36 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 36 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 34 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 32 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 31 C
# nvidia-smi -a |grep Gpu
        Gpu                         : N/A
        Gpu                         : 31 C

I read a lot about how people have tried make these things work. To make the GPU run/crunch data in a non GUI environment

Some uses “Wine”
Some uses OpenCL/CUDA
Most give up and uses the GUI to do it.

Here’s the proof from berkely.edu:
The jobs can be seen here. It’s CUDA55 which is the job being crunched by the GPU http://asteroidsathome.net/boinc/results.php?userid=2948
And here the jobs/work units is being handled http://asteroidsathome.net/boinc/host_app_versions.php?hostid=79453
Here you can see the active computers attached to the projects. It’s the computer named “halifax” with ID79453 http://asteroidsathome.net/boinc/hosts_user.php?sort=rpc_time&rev=0&show_all=0&userid=2948

If you got the time, please help me figure a way to make this headless ubuntu system which can and will use the GPU to crunch ;)

I read about a guy who worked with this issue for more than a year. I just don’t understand why the Boinc teams/Ubuntu developers hasn’t found a solution in between themselves!?!?

Kind Regards,
Dan

Hi,

I now tried to install minimum required X Server Packages:

sudo apt-get install xdm xorg

No difference ;(

Hi,

Gert-Jan ;)

I found a way which should have worked, but it didn’t. The guy who is coding the PPA for Ubuntu BOINC “FrancoCiancarlo” is trying to solve this. But they cant get it right. This is why I’ve tried with Ubuntu apt-get install boinc-client. And when that didn’t work, I tried with boinc directly from Berkely. This is the last attemp. Clearly there’s some issues here, which not even the top level of the guru’s can fix. Then how should I be able to do it. I’ve only been Linux’ing for little more than a year ;) + a little Redhat, SuSE and FreBSD. But this was a long time ago, and not very much.

Gert-Jan I’m thankfull for your help. I’ve written the top level guru’s at Ubuntu, I made a note at howtoforge and this is it for me.

If you got the time. Will you help me make an installation just like yours? I’ve been reading about debootstrap here https://help.ubuntu.com/lts/installation-guide/powerpc/linux-upgrade.html and I’m learning how to make the partitions manually. One of my old friends, a very talented Linux guy, once tried to teach me all this. So I have an idea of how it works. If less RAM, swap is important, he tried to teach me about the file systems too, FS2??? ReiserFS… I don’t remember much, but maybe I know it when I see it ;)
I’ll be reading about debootstrap and then if you got the time, will you then help me make a minimum installation like yours? I think its a lot smarter to make it that way instead of the way I do it. Of course it is! This is what I was looking for in the first place. It’s just that I’m most familiar with Ubuntu 12.04 commands etc. And these limited skills I still get to use, if we make a system like yours ;)

Kind Regards,
Dan

Here’s the 2 system I’m testing and the Rack I’m building for Boinc, SETI@ & Asteroid@