Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Notes on integrating the NVIDIA container runtime on Jetson platforms.

Layers required

In addition to the OE-Core and meta-tegra layers, you will need the meta-virtualization layer and the meta-oe, meta-networking, and meta-python layers from the meta-openembedded repository.

Configuration

  1. Add virtualization to your DISTRO_FEATURES setting.

Building

  1. Add nvidia-container-toolkit to your image to enable GPU-accelerated containers.

  2. See the NVIDIA Container Toolkit documentation for details on runtime configuration and usage.

  3. The Docker containers that NVIDIA supplies do not bundle most hardware-specific libraries, but expect them to be provided by the host OS. Be sure to include TensorRT, cuDNN, and/or other JetPack components in your image if you expect to run containers that need them.

  4. For containers that use GStreamer, include the Jetson-specific GStreamer plugins you may need. See Tegra-specific GStreamer plugins for the available plugin recipes.

  5. Consult the documentation in the branch of meta-virtualization you are using for information on how to configure Docker to register the nvidia runtime to be available at boot time, to avoid having to run the nvidia-ctk tool and restart the Docker service on every boot.