Closed michael-schaefer-eu closed 10 months ago
Hi @michael-schaefer-eu, have you set your default-docker runtime to nvidia like here?
https://github.com/dusty-nv/jetson-containers/blob/master/docs/setup.md#docker-default-runtime
Do you have CUDA Toolkit installed on your device, outside container? You should find that libcublas.so.10 file under /usr/local/cuda/lib64/libcublas
? On JetPack 4, CUDA/cuDNN/TensorRT are mounted from the host device.
Are you able to run pre-built l4t-pytorch container and import torch
? If not, the nvidia container runtime isn't working properly.
Hi @dusty-nv , thank you very much for your help.
Yes, docker runtime is set to nvidia:
sudo docker info | grep 'Default Runtime'
Default Runtime: nvidia
Cuda Toolkit is installed on the host:
ls -hal /usr/local/cuda/lib64/ | grep libcublas
lrwxrwxrwx 1 root root 17 Mar 1 2021 libcublasLt.so -> libcublasLt.so.10
lrwxrwxrwx 1 root root 25 Mar 1 2021 libcublasLt.so.10 -> libcublasLt.so.10.2.3.300
-rw-r--r-- 1 root root 33M Mar 1 2021 libcublasLt.so.10.2.3.300
-rw-r--r-- 1 root root 35M Mar 1 2021 libcublasLt_static.a
lrwxrwxrwx 1 root root 15 Mar 1 2021 libcublas.so -> libcublas.so.10
lrwxrwxrwx 1 root root 23 Mar 1 2021 libcublas.so.10 -> libcublas.so.10.2.3.300
-rw-r--r-- 1 root root 78M Mar 1 2021 libcublas.so.10.2.3.300
-rw-r--r-- 1 root root 93M Mar 1 2021 libcublas_static.a
Also OpenCV (if that matters):
python3 -c "import cv2; print('OpenCV version:', str(cv2.__version__));"
OpenCV version: 4.1.1
I will try out the torch container as soon as possible, I guess it's that one: dustynv/pytorch:1.10-r32.7.1
If the nvidia container runtime isn't working properly, can I fix that or do I have to reinstall the complete OS again?
Also /usr/local/cuda/include/
contains no cudnn headers, but I found it in /usr/include/cudnn.h
.
Is it possible that it has not been installed properly when I upgraded the system from L4T 32.5
to 32.7
?
Or is that OK that way?
Is there a script available to check all the required libs on the host? I couldn't find out anything.
If the container mounts aren't working on JetPack 4, typically I just reflash the system because they are tricky to debug/fix
thanks @dusty-nv ,
I re-flashed my jetson nano with the latest Jetpack 4.6.4 with L4T 32.7.4 (via the Docker SDK Manager, that alone took me a while to get it running ... )
The build process still failed with the same error as above, but this time at least the run
worked for me:
./run.sh dustynv/l4t-pytorch:r32.7.1
and
./run.sh dustynv/ros:iron-ros-base-l4t-r32.7.1
and both can load CV2 inside the container :
root@jbot:/# python3 -c "import cv2; print('OpenCV version:', str(cv2.__version__));"
OpenCV version: 4.5.0
I have a Jetson Nano on which I try to build the ROS Iron which fails.
Jetpack 4.6.4 L4T 32.7.4 Cuda 10.2.300 OpenCV 4.1.1
When I run
./build.sh ros:iron-ros-base
I get the following error when theopencv
image is built:From the logs:
logs/20231106_122338/build/opencv_r32.7.4-opencv.sh
logs/20231106_122338/build/opencv_r32.7.4-opencv.txt
Is it not possible to build ROS Iron on Jetson Nano or am I doing something wrong?