Docker and NVIDIA Container Toolkit Installation

The NVIDIA Container Toolkit is required in order to use GPUs inside Docker containers.

Install Docker and NVIDIA Container Toolkit:

• https://docs.docker.com/engine/installation/
  • Be sure to use Docker's repositories for the latest version of Docker.
• https://github.com/NVIDIA/nvidia-docker/blob/master/README.md

Building HEAVY.AI container

The Dockerfile assumes a copy of the HEAVY.AI tarball is in the same directory as the Dockerfile and is named heavyai-latest-Linux-x86_64.tar.gz.

To build the container, run:

mv heavyai-6.0.0-*-render.tar.gz heavyai-latest-Linux-x86_64.tar.gz
tar -xvf heavyai-latest-Linux-x86_64.tar.gz --strip-components=2 --no-anchored "docker/Dockerfile"
docker build .

where heavyai-6.0.0-*-render.tar.gz is the path to the HEAVY.AI tarball.

The container image id will be output on the last line of the build step. To assign a custom name and tag:

docker build -t heavyai/heavyai:v6.0.0 .

which will assign the name heavyai/heavyai and the tag v6.0.0 to the image.

Image layout

The data directory is at /var/lib/heavyai/storage.

The config file lives at /var/lib/heavyai/heavy.conf.

Running HEAVY.AI inside a container

docker run -d \
  --gpus all \
  -p 6273:6273 \
  --name heavyai \
  -v /path/to/storage:/var/lib/heavyai \
  heavyai/heavyai:v6.0.0

This starts the HEAVY.AI platform inside a container named heavyai, and exposes the Immerse visualization client on port 6273..

Data will be persisted to the host directory /path/to/storage.