Install Python+cuda+cudnn+tensorflow on WINDOW10

Software	Version
Window10	X64
Python	3.6.4 (64-bit)
CUDA	CUDA Toolkit 9.0 (Sept 2017)
CuDNN	CuDNN v7.0.5 (Dec 5), for CUDA 9.0

The above version of the test passed.

Installation steps:

1. to install python, remember to tick pip.

2. detects if CUDA is supported .

For more information on the NVIDIA website, see: Https://developer.nvidia.com/cuda-gpus, you can see if you can use TensorFlow with GPU support. Before installing and running TensorFlow, you need to install the Cuda driver first.

3. Installing CUDA Drivers

Visit Cuda's Download website: https://developer.nvidia.com/cuda-toolkit, you can see Cuda the latest version is 9.1 version, can be selected by the following "Legacy releases" Link to download the old version of the driver.

It is recommended that you choose a custom installation and then only Cuda. It is recommended to install the default file path.

4. Installing the CuDNN library

CUDNN Library (the NVIDIA CUDA?) Deepin Neural Network library) is a CUDA update package for deep neural networks that TensorFlow uses to accelerate deep learning on Nvidia GPUs. Can be downloaded from here, see: Https://developer.nvidia.com/cudnn.

To register an NVIDIA developer account first, it is free. After logging in, you will see various CUDNN downloads. This article downloads the Cuda 9.0, so it is determined that Cuda 9.0 has the Cudnn v7.0.5 selected.

The download is a ZIP file containing several folders, each containing a CUDNN file (a DLL, a header file, and a library file). Find your Cuda installation directory, here's what it should look like:

C:\Program Files\nvidia GPU Computing toolkit\cuda\v9.0

You can see the directory from the zip file also in this directory, that is, there is a bin, an include, a lib and so on. Copy the files from the zip to the relevant directory.

For example, drag the Cudnn64_7.dll file to the C:\Program files\nvidia GPU Computing toolkit\cuda\v9.0\bin directory, other similar.

5. Installing tensorflow

Install the nightly build version of TensorFlow. To install it from the command prompt, just enter: Pip install Tf-nightly-gpu

(The standard version of TensorFlow can also be installed.) Input: Pip install Tensorflow-gpu)

Once the installation is complete, enter in the command Prompt window: python

Open the Python editor, where you enter: Import TensorFlow

If the Cuda driver has an error, the Cudart64_xx.dll failure may be displayed, where XX is the version number.

If the Cuda driver is correct, but the CUDNN driver has an error, it may be shown that Cudnn64_x.dll is missing something, where X is a version number.

Install Python+cuda+cudnn+tensorflow on WINDOW10