Tensorflow GPU installation

Want to install , Look directly at the back TF install

1. Software to be installed
2. CUDA drive ,cuDNN,TF

CUDA drive ,cuDNN,TF Relationship is introduced

https://blog.csdn.net/xs11222211/article/details/84782046

except TF According to the latest version of the current period

1. Ubuntu edition Decide the highest Graphics card driver version
   Check the recommended graphics card driver version of the current system

   ubuntu-drivers devices 
   

2. Driver version Decide the highest cuda and cudnn edition
   Corresponding to the graphics card version cuda and cudnn edition
   https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html

3. cuda edition decision TF The highest version
   https://www.tensorflow.org/install/source#install_gpu_support_optional_linux_only
   

TF install

• Suggest use docker install
• stay tensorflow Of docker hub Find the direct version of the image to run
• You can also use the image provided cuda Such as the environment , Directly in conda Built in virtual environment TF

conda create --name tensorflow_gpu python=3.6
source activate tensorflow_gpu
conda install tensorflow-gpu==1.11.0

test TF Installation

python -c "import tensorflow as tf; print(tf.test.gpu_device_name())"

or

import tensorflow as tf
tf.test.is_gpu_available()