ImportError: /usr/lib/x86_64-linux-gnu/libcuda.so.1: file too short about personae HOT 5 CLOSED

nvidia-docker run -dit -p 8888:8888 -p 6006:6006 tensorflow/newbuild:2.0 /bin/bash
like me .you should use nvidia-docker to run

from personae.

@JS00000 You should use Nvidia-docker2.

from personae.

• macOS
  1. If the GPU of your mac is provided by NVIDIA, then you can use CUDA, and the version should be CUDA8, cuDNN6. Then you could run this project by the docker image mentioned in README.
  2. If the GPU of your mac is not provided by NVDIA, then you cannot use CUDA and tensorflow-gpu and the Docker image of this project, I did not mention it in README, sorry for that, for this case, you could only use tensorflow-cpu and conda to build the environment and run this project.

It's recommended to use docker on Ubuntu with NVIDIA GPU.

from personae.

It seems like your CUDA version or cuDNN version is not correct, can you tell me what version are they?

from personae.

I did nothing with CUDA or cuDNN in Docker. Just build the docker image using this project's Dockerfile.

Outside the Docker, I installed CUDA8.0 && CUDA9.1 in macOS, and CUDA9.1 is linked.

$ ls /Developer/NVIDIA
CUDA-8.0 CUDA-9.1

$ nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Tue_Dec_19_21:36:29_CST_2017
Cuda compilation tools, release 9.1, V9.1.128

And my cuDNN's version seems like is 5, but it is not work with CUDA9.1 now.

$ ls /usr/local/cuda/lib | grep cudnn
libcudnn.5.dylib
libcudnn.dylib
libcudnn_static.a

$ cat /usr/local/cuda/include/cudnn.h
cat: /usr/local/cuda/include/cudnn.h: No such file or directory

Should I install the right version of CUDA and cuDNN outside the Docker? Or install the CUDA8.0 and cuDNN6 in Docker manually? Thank you very much for explaining my confusion.

from personae.