I am looking for a way to install the CUDA toolkit without the use of a
.run installer from nvidia's website. I tried this before (and it installed successfully), however it caused havoc every time the kernel updated. I was advised by multiple sources to revert to an earlier backup and install CUDA again, using the package manager route.
So far, I have followed all the instructions at http://docs.nvidia.com/cuda/cuda-getting-started-guide-for-linux/#axzz3zypHweWG Chapter 3.6 - Ubuntu Package Manager Installation.
After several pre-installation checks, you are supposed to use the package manager to install a rather large .deb file from the nvidia website (or equivalently use a lightweight network-based .deb installer).
$ sudo dpkg -i cuda-repo-<distro>_<version>_<architecture>.deb
The documentation informs us that:
When using RPM or Deb, the downloaded package is a repository package. Such a package only informs the package manager where to find the actual installation packages, but will not install them.
Therefore, we complete the installation with the following:
$ sudo apt-get update $ sudo apt-get install cuda
For instance, I couldn't continue to setting the environment variables:
$ export PATH=/usr/local/cuda-7.5/bin:$PATH
/usr/local/cuda-7.5 was never created.
sudo apt-get purge cuda, I saw that some people had used
$ sudo apt-get install cuda-toolkit
with better results. Asking for the
nvcc -version command also suggests downloading this package.
This ran for a little while longer, which was reassuring. However after installation there was still no such directory as
Am I missing something obvious, or is this package manager method totally ineffectual?