To enable portability in Docker images that leverage GPUs, three methods of providing GPU support for Docker containers have been developed.
  • Native GPU support (preferred method if running Docker-ce 19.03 or later)
  • nvidia-docker2
  • nvidia-docker
Each of these methods mount the user mode components of the NVIDIA driver and the GPUs into the Docker container at launch. They allow NGC containers to take full advantage of NVIDIA GPUs.
  • DGX™ users should follow the instructions in the Preparing to use NVIDIA Containers Getting Started Guide.
  • NGC Cloud Images available from NGC Cloud Service Providers are set up with all the software needed to run NGC containers.
  • User running NGC containers on TITAN PCs, Quadro PCs, or vGPUs should follow the corresponding instructions.
  • Other users should follow the nvidia-docker installation documentation at nvidia-docker installation and install the latest NVIDIA Display driver for your GPU product type and series for your operating system. If NVIDIA drivers are not already configured on your system, then install them from here: Download Drivers.
  • Ensure you have an NVIDIA GPU supporting Compute Unified Device Architecture® (CUDA) version with compute capability 6.0.0 or higher. For example, Pascal GPU architecture generation or later.
  • Log into the NVIDIA® GPU Cloud (NGC) Container Registry located at using your NGC API key. For step-by-step instructions on how to gain access and get your API key, see NGC Getting Started Guide.
Note: HPC Visualization Containers have different prerequisites. For more information, see HPC Visualization Containers.