JustDEVblog

How to install Tensorflow 1.0.1 on Ubuntu 18 with CUDA 8.0

Steps to follow to remove all:

• sudo apt-get remove \-\-purge '^nvidia-.*'
• sudo apt-get remove \-\-purge '^libnvidia-.*'
• sudo apt-get remove \-\-purge '^cuda-.*'
• sudo rm -rf /usr/local/cuda*
• sudo apt update
• sudo apt upgrade
• sudo apt autoremove

Steps to install drivers

• sudo ubuntu-drivers autoinstall
• nvidia-smi

or

• $ nvidia-detector
• sudo apt install nvidia-driver-[your version]

Step to install cuda-toolkit

• go to ‘developer.nvidia.com/cuda-toolkit-archive’
• select ‘CUDA Toolkit 8.0 GA2’
• Next follow: Linux/x86/_64/Ubuntu/runfile(local) and download this file to your linux machine
• Install it: $ sudo sh cuda_8....linux.run, no drivers, no samples
  • accept
  • y
  • n
  • y
  • (/usr/local/cuda-8.0)
  • samplels: n
  • symbolic link: y

After that everything should work, you can check it by typing:

• conda create --name tf1-gpu-py27 python=2.7
• conda activate tf1-gpu-py27
• conda install pip
• ./miniconda3/envs/tf1-gpu-py27/bin/pip install tensorflow-gpu==1.0.1
• ./miniconda3/envs/tf1-gpu-py27/bin/python2
• alias python-2=/home/$USER/miniconda3/envs/tf1-gpu-py27/bin/python2
• python-2
  • >>> import tensorflow as tf
  • >>> tf.__version__ ‘1.0.1’
  • >>> tf.test.is_gpu_available()

If Error:
ImportError: libcudart.so.8.0: cannot open shared object file: No such file or directory

• sudo ldconfig /usr/local/cuda-8.0/lib64
• Add this to ~/.bashrc
  • export PATH=$PATH:/usr/local/cuda-8.0/bin
  • export LIBRARY_PATH=$LIBRARY_PATH:/usr/local/cuda-8.0/lib64
• run source ~/.bashrc