Missing cuda.csv, cudnn.csv, tensorrt.csv in /etc/nvidia-container-runtime/host-files-for-container.d/

Hi, I am trying to use CUDA and TensorRT inside the docker container with my Jetson Xavier with base image is nvcr.io/nvidia/l4t-tensorrt:r8.4.1-runtime. But it failed to mount Cuda and TensorRT to the container. I figured it out that it is missing files: cuda.csv, cudnn.csv, tensorrt.csv in /etc/nvidia-container-runtime/host-files-for-container.d/ :

$ ll /etc/nvidia-container-runtime/host-files-for-container.d/
total 24
drwxr-xr-x 2 root root  4096 Nov 25 06:48 ./
drwxr-xr-x 3 root root  4096 Jan 27 03:19 ../
-rw-r--r-- 1 root root 15870 Aug 25 14:38 l4t.csv

Can you help me how to mount CUDA and TensorRT correctly in the docker container or how to restore the missing csv files?
Thank you!

1 Like

I have the exact same issue. I’ve just flashed my Xavier Dev Kit with Jetpack 5.1, and can only see l4t.csv.

I’m missing all the nvidia-container-csv-* packages:

user@host:~$ dpkg --list | grep  nvidia-container-
ii  libnvidia-container-tools                  1.10.0-1                             arm64        NVIDIA container runtime library (command-line tools)
ri  nvidia-container-runtime                   3.9.0-1                              all          NVIDIA container runtime
ri  nvidia-container-toolkit                   1.11.0~rc.1-1                        arm64        NVIDIA Container toolkit

Weirdly they are not even available in the nvidia repo anymore for JetPack 5.1: Index

Has the way the cuda and tensorrt are mounted inside the docker runtime been changed in this version of JetPack?

I found the explanation to my problem in this thread: Host libraries for nvidia-container-runtime - #2 by dusty_nv

JetPack 5.x, only l4t.csv gets used (because CUDA/cuDNN/TensorRT/ect are installed inside the containers on JetPack 5 for portability). Does that answer your question?

This means you need to use l4t-tensorrt instead of l4t-base, or install the cuda and tensorrt deps inside manually.

Yep that is correct @magnusm, on JetPack 5 those libraries are installed inside the containers themselves.

If you want the CUDA and TensorRT development files/headers too, you can also try the l4t-jetpack container.

1 Like

Fyi, the l4t-jetpack container has not been released for JP 5.1

Sorry about that, we’re currently working on it - I did try running l4t-jetpack:r35.1.0 on JetPack 5.1 though, and it seemed to work (just some quick/basic checks I did)

No problem, thanks for the info 👍

Thank you @magnusm and @dusty_nv for your information. I solved my issue by using l4t-jetpack:r35.1.0

1 Like

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