WIN10 (64-bit) installing the TensorFlow GPU

"Python 3.6 + tensorflow GPU 1.4.0 + CUDA 8.0 + CuDNN 6.0"

There is no pycharm to install the Pycharm first.

1, python:https://www.python.org/downloads/release/python-364/

Pull to the bottom and select Windows x86-64 executable installer download.

Note the Add Python 3.6 to path check box, and then select Install Now.

2, TensorFlow GPU 1.4.0 in Pycharm settings--project interpreter to add the corresponding version, note that the PIP version can not be too high, Tested 9.0.1 feasible, and the latest 10.0.1 is not possible, as to how to reduce the PIP version, not yet known.

3, Cuda:https://developer.nvidia.com/cuda-toolkit-archive

Select the corresponding version, select Windows, x86_64, exe (local). During the installation process, you need to turn 360 off, and after the installation is complete, in the Windows Command Prompt window, enter: Nvcc-v Verify that the installation was successful.

4, cudnn:https://developer.nvidia.com/rdp/cudnn-archive#a-collapse712-91

Note that at this point the version of CUDNN is subject to cuda version limitations, and before the download also need to register or login to the NVIDIA account (I use QQ number login), it is recommended to use IE browser for this process.

After the download is complete, unzip and copy the files in its corresponding folder to the corresponding folder in the CUDA installation directory C:\Program files\nvidia GPU Computing Toolkit\cuda.

Test phase:

Enter the following program in Pycharm to test:

Import TensorFlow as TFH = Tf.constant ("hello,world! ") Sess=tf. Session () Print (Sess.run (h))

If you are writing a program that requires a GPU to run, simply add the following program to your code:

Config == = tf. Session (Config=config)

If the program error, most of the version is not on!!!

WIN10 (64-bit) installing the TensorFlow GPU