/ machine-learning

Installing TensorFlow GPU on Windows 10 and running image_retraining

This tutorial will show you how to get TensorFlow gpu running on your Windows 10 without Docker or Linux Bash.

I will be using** Nvidia GTX 1070**:

Nvidia GTX 1070

1. Get Python 3.6 installer

https://www.python.org/downloads/release/python-362/

2. Get CUDA 8

cuda_8.0.61_win10.exe – 2017 release

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

3. Get CUDNN (need to register as nvidia developer first)

cudnn-8.0-windows10-x64-v6.0.zip

https://developer.nvidia.com/rdp/cudnn-download

You should end up with this:

installing-tensorflow-gpu-windows-10-running-image_retraining_02

4. Install Python

5. Install CUDA 8

6. Extract cudnn to any location

(let’s say c:\programs\nvidia\cudnn)

7. Set Windows Path:

installing-tensorflow-gpu-windows-10-running-image_retraining_03

8. Create folder c:\tensorflow and open command line

9. Validate installation

10. Git clone or download zip of tensorflow

https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/image_retraining

Extract image_retraining directory

Place it in

c:\tensorflow\image_retraining

11. Use flowers example

Download https://www.tensorflow.org/versions/master/tutorials/image_retraining download flower images and place it in

c:\tensorflow\flower_photos

12. Run command:

installing-tensorflow-gpu-windows-10-running-image_retraining_04

If you by any chance get this error:

Try different CUDNN version.

As I remember well it was giving me this error while using cuDNN v7.0.2 (Sept 7, 2017), for CUDA 8.0 and changing it to cuDNN v6.0 (April 27, 2017), for CUDA 8.0 fixed the problem

To test your neural network run: