RuntimeError: Failed to dlopen libcuda.so

Accelerated Computing CUDA CUDA Setup and Installation
, ,
1

Hi all,

I am using Nvidia Jetson Xavier NX device and trying to import cudf and cuml libraries in RAPIDS Jupyter notebook docker image. But when executing the notebook cell, I am getting an error as below.
RuntimeError: Failed to dlopen libcuda.so

IMG-0835
IMG-08351225×428 82 KB

libcuda.so is available in the “/usr/local/cuda/targets/sbsa-linux/lib/stubs:” and I have set the LD_LIBRARY_PATH into this folder path as well. But the error is still coming up. Do I need to set any other system paths apart from the LD_LIBRARY_PATH?

Any advice would be greatly appreciated!

Thank you

2

libcuda.so in the stubs directory is a library placeholder that contains no actual code. It is used for linking.

libcuda.so is part of NVIDIA drivers. You either have no drivers installed or you haven’t exposed the GPU to the docker container (docker run --gpus all ... the option is requires nvidia-container-runtime).

2 Likes
3

Thank you so much! You have saved a lot of time to me.