Using repositories without opengl files

Accelerated Computing CUDA CUDA Setup and Installation
#1

Hi,

I have an issue with the fedora 19 x64 repository dependencies, that I have not been able to solve :-/

I am trying to install the cuda framework and driver through the repository, without installing the opengl files. The reason that I do not want opengl files is that I want to use my GTX TITAN for computation only, and use the onboard intel GPU for my display.

However, I have not been able to find the repository package that provides the opengl files so that I can remove them, without also removing the cuda driver(and parts of the cuda tollkit).

As such my current solution is to install the cuda tools(but not driver) through the repository, and then install(and for each new kernel reinstall) the driver through the run file with the --no-opengl-files flag. However this is not a very clean solution, and I cannot believe that I am the only one using my integrated gpu for display and the CUDA gpu for computation. Have anybody solved this problem? And if not, does anybody know if there is a feature request option somewhere, so that it might be integrated in the next version?

#2

Hi,

Any update on this issue? I don’t see a CUDA 6.5 repo rpm for CentOS/RHEL 7, and I have to use the procedure mentioned by kennethsb when using discrete graphics for display.

Thanks in advance!

#3

CentOS/RHEL 7 is not an officially supported distro for CUDA 6.5

#4

RHEL/CentOS 7 is now supported officially in the latest CUDA 7.0 Release Candidate.

For using repositories without opengl files:
The DEB installers don’t provide this option as they integrate nicely with bumblebee.
In the RPM installers, it’s possible for users install a combination of the driver packages that does not provide the GL files. [Attaching the instructions for reference]

RHEL6.6:
- yum install xorg-x11-drv-nvidia-libs nvidia-uvm-kmod

Note, user may need the removal of  xorg.conf at /etc/X11/ if using a third-party display GPU after installation to be able to boot to X screen.

RHEL7.0:
- yum install xorg-x11-drv-nvidia-libs nvidia-uvm-kmod

Notes,
-  user may need the removal of  xorg.conf at /etc/X11/ and libglx.so at /usr/lib64/nvidia/xorg/ if using a third-party display GPU after installation to be able to boot to X screen.
-  user may need the removal of “nomodeset” parameter at kernel boot up command line if using a third-party display GPU to resolve the X screen flickering symptom.

Fedora21:
- yum install xorg-x11-drv-nvidia-cuda akmod-nvidia

OpenSUSE13.2:
- zypper install x11-video-nvidia-G03 nvidia-uvm-gfxG03-kmp-desktop

SLES12:
- zypper install x11-video-nvidia-G03 nvidia-uvm-gfxG03-kmp-default

Hoping it helps.

#5

Thank you, this works great, also with the final release of cuda 7(now running opensuse since the fedora support is not there yet)!