Install tensorrt in deepstream container

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU) Jetson Orin Nano Developer Kit
• DeepStream Version 7.1
• Issue Type( questions, new requirements, bugs) Question

I am having issues using tensorrt on my Jetson Orin Nano Devkit. I flashed it using SDK Manager, with jetson linux 36.4, and Jetpack 6.1 and have been using the docker image nvcr.io/deepstream:7.1-samples-multiarch.

I can run the sample applications, and now I want to run my own models. As I try to build the models, I get the error:

fatal error: NvInfer.h: No such file or directory

The file exists on my host system under /usr/include/aarch64-linux-gnu but it can’t be found in the container.

pip list shows that I have tensorrt 10.3.0 installed in the container. apt search tensorrrt | grep tensorrt however shows only:

tensorrt-libs/now 10.3.0.26-1+cuda12.5 arm64 [installed, local]

Natively on Jetson, I can apt install tensorrt, but I cannot inside the docker container, it gives me:

Unable to locate package tensorrt

When trying to do apt update I get:

W: Failed to fetch http://l4t-repo.nvidia.com/common/dists/36.4/InRelease Something wicked happened resolving ‘l4t-repo.nvidia.com:80’ (-5 - No address associated with hostname)

Do you know what I could do to solve this?

5

If you have already installed Jetpack-6.1 correctly. Try the following command line.

docker run -it --rm --net=host --runtime nvidia  -e DISPLAY=$DISPLAY -w /opt/nvidia/deepstream/deepstream-7.1 -v /tmp/.X11-unix/:/tmp/.X11-unix nvcr.io/nvidia/deepstream:7.1-triton-multiarch
6

Thank you!

I get another error instead, still related to tensorrt;

‘class nvinfer1::IBuilder’ has no member names ‘buildEngineWithConfig’

but it’s related to tensorrt version 10.3.0.26 so that makes sense.

However, when I run the triton-multiarch-container I also get this:

ERROR: This container was built for NVIDIA Driver Release 560.28 or later, but version 540.4.0 was detected and compatibility mode is UNAVAILABLE.

It doesn’t sound good, any idea why this is the case? As far as I’ve understood, the drivers are included when flashing Jetson Linux? And thus should be compatible?

7

Some TensorRT APIs have been removed in 10.3, please refer to this link to migrate your code

This is a known issue. The driver in Jetpack is an old version and needs to be updated later. You can ignore this warning for the moment.

8

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks