Minimize NGC l4t-tensorrt runtime docker image

Hi,

I’m looking at optimising our application dockerfile based on the Tensorrt image. I saw similar posts in the past such as (Tensorrt minimum runtime on docker - #6 by 1342868324) but there wasn’t an answer for it.

Is it possible to create a much smaller runtime container for the NGC l4t-tensorrt runtime docker images? Specifically, the r8.6.2-runtime docker image. Compared to the r10.3.0-runtime the difference is around 2.11GB (4.92 GB vs 2.81GB).

Hi,

You can try to build an l4t-tensorrt from the l4t-cuda base.
This allows you to install the necessary libraries only and reduce the image size.

The base image is available here:

And you can find the corresponding JetPack components below:
https://repo.download.nvidia.com/jetson

Thanks.

1 Like

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.