After a deep learning VM is deployed in VMware Private AI Foundation with NVIDIA, the NVIDIA vGPU guest driver status is unlicensed.

Problem

The /var/log/vgpu-install.log file contains one of the following messages:

License Status: Unlicensed
Unlicensed (Restricted)

Cause

The NVIDIA vGPU client configuration token that you pass as a value to the vgpu-license OVF property or to the catalog setup wizard for private AI in VMware Aria Automation is invalid, expired, or incorrectly formatted.

Solution

  • Verify the validity of the client configuration token.

  • Verify that the vGPU license is correctly formatted and follows the JWT token format, which typically looks like eyxxxx.eyxxxxx.xxxxx.

    You can decode the JWT token at jwt.io to check the expiration date and node server URL.

  • The vGPU license token also saved in /etc/nvidia/ClientConfigToken/client_configuration_token.tok.

  • To troubleshoot the problem further, run this command to check for specific error messages related to the communication to the NVIDIA license server.
    cat /var/log/syslog | grep -i nvidia
To apply a new token, following these steps:
  1. Replace the content of the /etc/nvidia/ClientConfigToken/client_configuration_token.tok file with a new token, run the following command:
    echo -n $vgpu_license_token > /etc/nvidia/ClientConfigToken/client_configuration_token.tok
  2. Restart the NVIDIA service.
    /etc/init.d/nvidia-gridd restart
    
  3. Verify the license status of the NVIDIA vGPU guest driver.
    nvidia-smi -q | grep -i "license status" | sed 's/^[ \t]*//'