ImportError: libcublas.so.10.0: cannot open shared object file: No such file or directory

Question

I have installed Cuda 10.1 and cudnn on Ubuntu 18.04 and it seems to be installed properly as type nvcc and nvidia-smi, I get proper response:

user:~$ nvcc -V         


        

Answer 1

The problem is caused by your current cuda version which is 10.1 (as we can see from the top right corner of your image).

As you can see from the official TF website, the correspondence between tf and cuda is: TF website for the chart

Version                 cuDNN    CUDA
tensorflow-2.1.0         7.6       10.1
tensorflow-2.0.0         7.4       10.0
tensorflow_gpu-1.14.0    7.4       10.0
tensorflow_gpu-1.13.1    7.4       10.0

Thus, you can either upgrade your tf to 2.1 or downgrade your cuda with:

conda install cudatoolkit=10.0.130

Then it would automatically downgrade your cudnn as well.

Answer 2

If using Cuda 10.1 (as directed in https://www.tensorflow.org/install/gpu), the problem is that libcublas.so.10 was moved out of the cuda-10.1 directory and into cuda-10.2(!)

Copying from this answer: https://github.com/tensorflow/tensorflow/issues/26182#issuecomment-684993950

... libcublas.so.10 sits in /usr/local/cuda-10.2/lib64 (surprise from nvidia - installation of 10.1 installs some 10.2 stuff) but only /usr/local/cuda is in include path which points to /usr/local/cuda-10.1.

The fix is to add it to your include path:

export LD_LIBRARY_PATH=/usr/local/cuda-10.2/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

Note: This fix is known to work in Cuda 10.1, V10.1.243 (print your version with nvcc -V).

Answer 3

Change my tensorflow version solved my problem.

check this issue 1https://github.com/tensorflow/tensorflow/issues/26182)

Official tensorflow-gpu binaries (the one downloaded by pip or conda) are built with cuda 9.0, cudnn 7 since TF 1.5, and cuda 10.0, cudnn 7 since TF 1.13. These are written in the release notes. You have to use the matching version of cuda if using the official binaries.

Answer 4

In case someone is still having this issue, libcublas.so.10 can exist but under the name libcublas.so.10.0

So, you can fix it by running:

sudo ln libcublas.so.10.0.130 libcublas.so.10

in /usr/local/cuda-10.0/lib64

Answer 5

I faced similar issue when trying to install spconv.

File "/home/kmario23/anaconda3/envs/py38/lib/python3.8/site-packages/torch/_ops.py", line 105, in load_library
    ctypes.CDLL(path)
  File "/home/kmario23/anaconda3/envs/py38/lib/python3.8/ctypes/__init__.py", line 373, in __init__
    self._handle = _dlopen(self._name, mode)
OSError: libcublas.so.10: cannot open shared object file: No such file or directory

Installing the cuda toolkit version 10.1 inside the specific environment resolved the issue:

$ conda install -c anaconda cudatoolkit=10.1

Answer 6

I had the same issue. I fixed it by adding the below command to the '.bashrc' file.

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-10.0/lib64/

System configuration:

Ubuntu 16.04 LTS
Tensorflow GPU 2.0beta1
Cuda 10.0
cuDNN 7.6.0 for Cuda 10.0

I used conda to configure my system.