You will require to update your
LD_LIBRARY_PATH, hence that it points to the
/usr/local/cuda-9.0/lib64. Include the following line to your
.bashrc file (or any other terminal you usage)
The Tensorflow Binaries do not support Cuda 9.1 as of currently. Hence, the resolution might be only installing Tensorflow from source.
I too had the above mentioned problem at the time I had installed Cuda 9.1+cuDNN 6 for Tensorflow and this is what performed for me.
Moreover, you have the following two options:
CUDA 8.0 + Tensorflow - the most stable release of CUDA that performs with Tensorflow Binaries.
CUDA 9.0 + Tensorflow - current CUDA version that Tensorflow Binaries are compatible with.
Attempt the following code
pip3 install --upgrade tensorflow-gpu==1.4
Afterwards you typing this command
pip3 install --upgrade tensorflow-gpu==1.4 in the terminal, the tensorflow will downgrade to 1.4.0. This bug happened by tensorflow 1.6.0.
Tensorflow version >= 1.5 requires CUDA version > 8.0. Hence, in case you have CUDA version 8.0, you can downgrade your tensorflow version to 1.4.
pip install tensorflow-gpu==1.4
You're attempting to make
tensorflow-gpu 1.4, which is too old to be compatible with CUDA 9.1.
You can attempt to make a symbolic link to force the system to use CUDA 9.1, though it's not sure it'll succeed
You can follow this tutorial which explains how to easily install CUDA 9.0 on Ubuntu 16.04.
In case it still doesn't perform and that you trully need to attempt to install Tensorflow with CUDA 9.1, you can attempt to install CUDA, CUDNN and NCCL sources from the NVIDIA site, and then simulate the way these packages would have been installed from ppa.
At one time you've downloaded the
tar packages, type the following to configure CUDA:
sudo mkdir -p /usr/local/cuda /usr/local/cuda/extras/CUPTI /usr/local/cuda/nvvm
sudo ln -s /usr/bin /usr/local/cuda/bin
sudo ln -s /usr/include /usr/local/cuda/include
sudo ln -s /usr/lib/x86_64-linux-gnu /usr/local/cuda/lib64
sudo ln -s /usr/local/cuda/lib64 /usr/local/cuda/lib
sudo ln -s /usr/include /usr/local/cuda/extras/CUPTI/include
sudo ln -s /usr/lib/x86_64-linux-gnu /usr/local/cuda/extras/CUPTI/lib64
sudo ln -s /usr/lib/nvidia-cuda-toolkit/libdevice /usr/local/cuda/nvvm/libdevice
Thereafter you can either download cuDNN and NCCL from source and configure them in the similar fashion as above , or download their
.deb package and view in case the installation performs currently. In case it doesn't, then attempt to install from source.
There is an other problem concerned to this. In case you've installed Cuda 9.1 or 9.0 in your lib64 folder you will view that the links come from another folders and it can have a file like this libcublas. Hence,.9 which is originally the similar file without .0 part. I think you can only create another link to libcublas.so.9. -> libcublas.so.9.0.
The issue is concerned to the path of intealled cuda. You require to tell where the files are located by including global environment variable.
LD_LIBRARY_PATH = /usr/local/cuda/lib64:$LD_LIBRARY_PATH
You can include this into your .bash_profile to let it run globaly.