NGC iGPU docker not working on Jetson Thor

[quan.luo.101]

I’m using the 25.10-py3-igpu ( TensorRT | NVIDIA NGC ) docker to run some application on Jetson Thor.

My docker command: docker run -it --gpus all --runtime nvidia --network host --privileged -v /home/user:/app nvcr.io/nvidia/tensorrt:25.10-py3-igpu

Jetpack 7.0 has already been installed as well.

Then I tried to run the mnist example provided in the webpage but failed with

/usr/bin/ld: warning: libnvdla_compiler.so, needed by /usr/lib/gcc/aarch64-linux-gnu/13/../../../aarch64-linux-gnu/libnvinfer.so, not found (try using -rpath or -rpath-link)
/usr/bin/ld: /usr/lib/gcc/aarch64-linux-gnu/13/../../../aarch64-linux-gnu/libnvinfer.so: undefined reference to `nvdla::createWisdom()'

And I cannot use tensorrt to build my own engine as well.

If I use the 25.10-py3 GPU docker, everything works fine. Why is that? Shouldn’t we use iGPU docker on Jetson?

[AastaLLL]

Hi,

Thanks for your testing.

Thor uses the SBSA driver, so it can use the standard container that is built for an ARM system.
The igpu container is a customized container for the device (ex, Orin, Xavier) that uses nvgpu driver.

Thanks.