Installing the Linux vGPU driver and configuring the licensing token

After completing this instruction, the vGPU driver will be installed and the licensing token will be configured.

The vGPU driver does not support updates. To change its version, first remove it, and then install it again by repeating the steps in this instruction.

1. Check your VM's configuration type:

   1. Go to your VK Cloud personal account.
   2. Go to Cloud Computing → Virtual Machines.
   3. In the list, find the VM for which you need to install the vGPU driver. Make sure the Type column shows a vGPU configuration flavor.

2. Connect the VM to an external network if you haven't done so already.

3. Connect to the VM with vGPU.

1. Check whether the vGPU driver is installed:

    
   nvidia-smi --query-gpu=driver_version,virtualization.mode,virtualization.virtualized,name --format=csv,noheader

   The following results are possible:

   • A No such file or directory error means the driver is not installed.
   • The command output shows a driver version, and the virtualization type is shown as vGPU. This means the driver is already installed and the NVIDIA stack was initialized successfully. No further action is required.
   • nvidia-smi runs, but the output shows a ... couldn't communicate with the NVIDIA driver error. This means the driver is damaged or an incorrect version is used (for example, for dedicated GPUs). You need to reinstall the driver: first remove it, then install it again by repeating the steps in this instruction.

2. Prepare the OS:

   Ubuntu/Debian

   Red Hat Enterprise Linux

   1. (Optional) Update the system components and reboot the VM to avoid compatibility issues.

      Run the commands:

       
      sudo apt update &&sudo apt upgrade -y &&sudo reboot

   2. Install the additional packages required to install the driver:

       
      sudo apt install -y build-essential dkms perl pkg-config libelf-dev linux-headers-$(uname -r)

   1. (Optional) Update the system components and reboot the VM to avoid compatibility issues.

      Run the commands:

       
      sudo dnf update -y &&sudo reboot

   2. Install the additional packages required to install the driver:

       
      sudo dnf -y install gcc make perl dkms pkgconf-pkg-config elfutils-libelf-devel "kernel-devel-uname-r == $(uname -r)" "kernel-headers-uname-r == $(uname -r)"

3. Download the NVIDIA® GRID driver using one of the following methods:

   wget

   curl

    
   wget https://hub.mcs.mail.ru/repository/gpu-drivers-raw/bin/nvidia/guest-drivers/latest/vgpu_driver_linux.run

    
   curl -O https://hub.mcs.mail.ru/repository/gpu-drivers-raw/bin/nvidia/guest-drivers/latest/vgpu_driver_linux.run

4. Grant execute permission to the downloaded file:

    
   chmod +x vgpu_driver_linux.run

5. Run the file and follow the installer's instructions:

    
   sudo ./vgpu_driver_linux.run

6. Reboot the VM.

1. Download the licensing script using one of the following methods:

   wget

   curl

    
   wget https://hub.mcs.mail.ru/repository/gpu-drivers-raw/bin/nvidia/nvidia-token-fetcher/latest/nvidia_token_fetcher

    
   curl -O https://hub.mcs.mail.ru/repository/gpu-drivers-raw/bin/nvidia/nvidia-token-fetcher/latest/nvidia_token_fetcher

2. Grant execute permission to the script:

    
   chmod +x nvidia_token_fetcher

3. Run the script:

    
   sudo ./nvidia_token_fetcher

4. Check the licensing configuration:

   1. Make sure the licensing token file exists:

       
      sudo ls /etc/nvidia/ClientConfigToken/

   2. Check the license status:

       
      nvidia-smi -q | grep -i license

      Output for a correctly configured license:

       
      vGPU Software Licensed Product    License Status: Licensed

5. (Optional) Configure automatic licensing token verification at system startup.

   If the VM was created from a prepared OpenStack image, automation can be configured using cloud-init. You can also use systemd.

   cloud-init integration

   Creating a systemd unit file

   Copy the script file to a dedicated directory:

    
   sudo cp nvidia_token_fetcher /var/lib/cloud/scripts/per-boot/

   1. Save the script file to a shared directory:

       
      sudo cp nvidia_token_fetcher /usr/local/bin/

   2. Create a new unit file for systemd. Example using the nano text editor:

       
      sudo nano /etc/systemd/system/nvidia-token.service

      You can use any other editor instead of nano, for example vi.

   3. Copy the following content into the file and save it:

       
      [Unit]Description=NVIDIA Licensing ScriptAfter=network.target[Service]Type=oneshotExecStart=/usr/local/bin/nvidia_token_fetcherTimeoutStartSec=15RemainAfterExit=false[Install]WantedBy=multi-user.target

   4. Run the command to enable automatic startup on system boot:

       
      sudo systemctl enable nvidia-token.service