bhavsarpratik / install_Tensorflow_Windows

Installation instructions for Tensorflow GPU on Windows

Tensorflow CPU/GPU installation on Windows 10 64bit

Easiest method

• Install Anaconda
• Run conda install tensorflow-gpu (This will take care of all dependency installations - Nvidia toolkit, cuda, visual c++ and python library

Long method

• If you are installing Tensorflow GPU version, check if your NVIDIA GPU is supported for Tensorflow and has Compute Capability >= 3.0
• As on 24/3/2017 Tensorflow is supported only on 2.7.x and 3.5.x. So make sure you have this version Python 64bit installed
• Add Python directory to your environment variable path after installation

Tensorflow installation steps

python -m pip install tensorflow-gpu # for Tensorflow **GPU** installation
python -m pip install tensorflow # for Tensorflow **CPU** installation

• Put the tensorflowvisu.mplstyle in C:\Users\Pratik\AppData\Local\Programs\Python\Python35\Lib\site-packages\matplotlib\mpl-data\stylelib
• Install visual studio community edition
• Install NVIDIA Cuda
• Download cuDNN and put those files where CUDA was installed (merge folders)
  • cudnn64_5.dll (cuda\bin\cudnn64_5.dll) from the zip archive into C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin\
  • cuda\include\cudnn.h to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include\
  • cuda\lib\x64\cudnn.lib to C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\lib\x64\
  • Put C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\bin to environment variable path

Done!!

Check installation with these lines

# Below code print out active GPUs   

hello = tf.constant('Hello, TensorFlow!')    
sess = tf.Session()    
a = tf.constant(10)    
b = tf.constant(32)    
print(sess.run(hello))   
print(sess.run(a + b))    

from tensorflow.python.client import device_lib

def get_available_gpus():
    local_device_protos = device_lib.list_local_devices()
    return [x.name for x in local_device_protos if x.device_type == 'GPU']

print(get_available_gpus()) 

Force using CPU instead of GPU

with tf.device('/cpu:0'):
  a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
  b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
  c = tf.matmul(a, b)
  
# Creates a session with log_device_placement set to True

sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
print sess.run(c)