Install Nvidia graphics card driver on Ubuntu along with cuda libraries

Home > Ubuntu > Server or Desktop administration > Install Nvidia graphics card driver on Ubuntu along with cuda libraries

1. First determine the version of Ubuntu using:

   cat /etc/*release

2. Then determine the Nvidia card installed on machine using one of the below methods:
   1. Using lspci

      sudo  lspci | grep -i --color 'vga\|3d\|2d'

   2. To to System > Preferences > Hardware information and check card information
   3. Install and use hardinfo to get information about card

      sudo apt-get install hardinfo
      hardinfo

   Refer https://www.cyberciti.biz/faq/linux-tell-which-graphics-vga-card-installed/

3. Download drivers from nvidia site ( http://www.nvidia.com/download/driverResults.aspx/118962/en-us )

   Example driver download link http://us.download.nvidia.com/tesla/375.66/nvidia-diag-driver-local-repo-ubuntu1604_375.66-1_amd64.deb

4. Install the downloaded driver, update and upgrade OS, Install cuda drivers using:

   dpkg -i <driver>.deb
   apt-get update
   apt-get upgrade
   apt-get install cuda-drivers

5. In some cases before installing cuda drivers we may have to setup cuda-keyring using:
   1. Install required kernel headers:

      sudo apt-get install linux-headers-$(uname -r)

   2. Download required keyring deb file and install using:

      distribution=$(. /etc/os-release;echo $ID$VERSION_ID | sed -e 's/\.//g')
      wget https://developer.download.nvidia.com/compute/cuda/repos/$distribution/x86_64/cuda-keyring_1.0-1_all.deb
      sudo dpkg -i cuda-keyring_1.0-1_all.deb

   Refer https://docs.nvidia.com/datacenter/tesla/tesla-installation-notes/index.html.

6. Reboot the machine and test
7. After reboot if screen is blank edit /etc/default/grub and append nomodeset option to GRUB_CMDLINE_LINUX_DEFAULT. Do not remove existing options append with space bar.

   GRUB_CMDLINE_LINUX_DEFAULT="nomodeset"

8. Run below and verify that output contains information about installed graphics card.

   /usr/local/cuda/bin/nvcc --version
   nvidia-smi

   Refer https://stackoverflow.com/questions/9727688/how-to-get-the-cuda-version

9. If setup is not working then as per https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html there is combination of OS, GCC, nvidia drivers, etc. that is tested for particular OS. We can validate those via

   uname -a
   gcc --version
   ldd --version
   nvidia-smi
   /usr/local/cuda/bin/nvcc --version

10. If required perform following additional steps

    sudo apt-get install nvidia-gds
    sudo apt-get install cuda-drivers-515
    conda install cuda -c nvidia
    cat /proc/driver/nvidia/version

11. Execute below in current shell to set PATH and LD_LIBRARY_PATH properly for current shell

    export PATH=/usr/local/cuda-11.7/bin${PATH:+:${PATH}}
    export LD_LIBRARY_PATH=/usr/local/cuda-11.7/lib${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

    Ideally we should use "module load", "module available" etc. to set appropriate environment for required cuda version based on the programs being compiled / executed.

12. Edit /etc/ld.so.conf and append a new line:

    /usr/local/cuda-11.7/lib64

13. Update ldconfig using

    sudo ldconfig

14. We can further test the installation with cuda samples from https://github.com/nvidia/cuda-samples. For that use

    sudo apt-get -y install git
    git clone https://github.com/NVIDIA/cuda-samples.git
    cd cuda-samples
    make

    All tests and compilations succeed except for 0_Introduction/simpleMPI perhaps due to missing MPI libaries. We did 'rm -rf 0_Introduction/simpleMPI' to delete that sample and compile all others using make

15. Especially, we ran Samples/1_Utilities/deviceQuery from below path

    ./bin/x86_64/linux/release/deviceQuery

16. Note after this we tried to run python based code from https://www.geeksforgeeks.org/running-python-script-on-gpu/ It did not work. Perhaps we have not installed python-cuda libraries to make the code work

Home > Ubuntu > Server or Desktop administration > Install Nvidia graphics card driver on Ubuntu along with cuda libraries