Smaller docker image than l4t-base for Jetson Nano to use in production?

I want to distribute applications that run on Jetson Nano as docker images.
Each Jetson Nano device has an agent program installed and it pulls and runs an application docker image from the Internet.

l4t-base image is convenient while development but it is too large (~ 1GB) for remote installation.

Are there any small docker images (less than 100MB, for example) that include runtime for Jetson Nano?

Applications might use:

  • CUDA
  • TensorRT
  • cuDNN

Hi,

Sorry that there is no docker image can meet you requirement.

There is two public docker image for Jetson device currently:
nvidia:l4t-base: 402.71 MB with CUDA toolkit installed only.
nvidia:deepstream-l4t: 757.44 MB with CUDA, cuDNN, TensorRT and Deepstream installed.

Thanks.

Hi,

nvidia:l4t-base : 402.71 MB with CUDA toolkit installed only.

402.71MB is not that bad for my use. However it seems 1.03GB to me.

% docker pull nvcr.io/nvidia/l4t-base:r32.3.1
% docker images |grep l4t
nvcr.io/nvidia/l4t-base       r32.3.1           aaaa63e7b12d        4 months ago        1.03GB

What’s wrong?

Thanks,

Hi,

Sorry for the missing.

The 402.71MB should be the compressed size.
https://ngc.nvidia.com/catalog/containers/nvidia:l4t-base/tags

Currently, our docker image doesn’t include CUDA library but mounting it from the host volume.
This is because CUDA is too large to be included into the image.
We are trying to solve it by extracting CUDA toolkit essential part.

Currently, l4t-base:r32.3.1 is our minimal image.
Once a new image becomes available, we will update a comment here.

Thanks.

1 Like