Installing TensorFlow with GPU on Windows

Basic Environment

OS Vision: Windows 7

CPU: Intel Core 2 Extreme QX6850

GPU: NVIDIA GeForce GTX TITAN X

RAM: DDR2 8G

Python 3.5.4

Download: https://www.python.org/downloads/release/python-354/

Add %PATH% — (How to add path: Click)

C:\Python\Python35; 
C:\Python\Python35\Scripts;

CUDA8.0 GA2

Download: https://developer.nvidia.com/cuda-toolkit-archive

Add %PATH%

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin; 
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\libnvvp;

cuDNN 5.1

Download: https://developer.nvidia.com/cudnn

Add %PATH

C:\cuda\bin;

Anaconda 3

Download: https://anaconda.org/

Add %PATH%

C:\ProgramData\Anaconda3; 
C:\ProgramData\Anaconda3\Scripts;

CMD Setting

Install TensorFlow-gpu

C:> pip3 install --upgrade tensorflow-gpu 
C:> conda create -n tensorflow-gpu python=3.5 
C:> activate tensorflow-gpu 
(tensorflow-gpu)C:> pip install --ignore-installed --upgrade tensorflow-gpu

Install Keras

(tensorflow-gpu)C:> conda install keras

Install Jupyter notebook with GPU

(tensorflow-gpu)C:> conda install ipython 
(tensorflow-gpu)C:> pip install jupyter

Leave virtual environment

(tensorflow-gpu)C:> deactivate

Using GPUs

1. MNIST_CNN

https://github.com/fchollet/keras/blob/master/examples/mnist_cnn.py

Time evaluation

GPU: around 15s per epoch

CPU: around 400s per epoch

2. Others

Using GPUs | TensorFlow

Reference

https://www.tensorflow.org/install/install_windows