If Jetson Nano comes with an SDK and libraries, what is the need of Docker then?

Jetson Nano comes with an SDK which has a lot of libraries. What is the need to install and set up a Docker container (it’s recommended in their tutorials and inference demos)? Isn’t it that Docker becomes an isolated environment or can it still interface with the tools installed on Nano’s system?

PS pardon me if it’s in the wrong topic, but it’s the closest fit I found


This is the vGPU forum. I am going to move your topic over to the Jetson Nano category for better visibility.


SDK manager will set up the OS and some basic libraries, ex. CUDA, cuDNN, …, etc.
And docker usually is used for specific usage, ex. TensorFlow, PyTorch, …, etc.
(This won’t be installed by default due to the limited storage of Jetson)

Our tutorial may provide a docker container with all the dependencies installed.
So the user doesn’t need to do this manually.