Step 1 of 8

1. System Requirements

Confirm your hardware meets the minimum requirements. Follow the instructions based on your GPU type.
Tip: Ensure your OS and GPU drivers are up to date. 
# Check NVIDIA drivers and CUDA toolkit
nvidia-smi
nvcc --version
  • NVIDIA Driver ≥ 450.80.02
  • CUDA Toolkit ≥ 11.0
# Verify GPU details
lspci | grep -i vga
glxinfo | grep "OpenGL version"
  • Up-to-date GPU drivers
  • Vulkan/Mesa support enabled

2. Docker Installation

Confirm Docker is installed and updated. This step ensures your container runtime is ready. 
# Check Docker version
docker --version
If Docker is not installed, please follow the official Docker installation guide.

3. Docker Configuration

Configure Docker for GPU support. Follow the instructions for your selected GPU type. 
# Install NVIDIA Container Toolkit
sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker
# Install Mesa Utils for GPU access
sudo apt-get install -y mesa-utils
sudo usermod -aG video $USER

4. NGC API Key Generation & Usage

For NVIDIA users, generate an API key from your NGC account:

To generate your API key:
  1. Log in to your NGC account at ngc.nvidia.com.
  2. Go to your account settings and find the API Key section.
  3. Click Generate API Key and copy the key provided.

Using your API key: First, set your API key as an environment variable, then log in using it. 
# Set your NGC API key as an environment variable
export NGC_API_KEY=
# Log in to the NVIDIA container registry using your API key
docker login nvcr.io -u '$oauthtoken' -p $NGC_API_KEY

5. Container Setup & Pulling NGC Packages

For NVIDIA users, pull specific NGC packages (e.g., TensorRT). For Non‑NVIDIA users, pull a generic base image. 
# Pull a specific NGC package (e.g., TensorRT)
docker pull nvcr.io/nvidia/tensorrt:21.06-py3
# Pull a base image from Docker Hub
docker pull ubuntu:latest
# Run a container with GPU access
docker run -it --device=/dev/dri:/dev/dri [IMAGE] [COMMAND]

6. Verification

Confirm the container can communicate with the GPU. 
# Run NVIDIA verification
docker run --gpus all nvidia/cuda:11.0-base nvidia-smi
# Run OpenCL verification
docker run --device=/dev/dri ubuntu clinfo

7. Troubleshooting

Review common issues and apply fixes if needed. 
# Restart Docker if needed
sudo systemctl restart docker

8. Final Review & Cleanup

Review the installation and run cleanup commands to free resources. 
# Clean up unused Docker resources
docker system prune -f
docker image ls
docker container ls -a