Skip to content

Latest commit

 

History

History
85 lines (63 loc) · 4.15 KB

Installing_Tensorflow_GPU_(Windows).md

File metadata and controls

85 lines (63 loc) · 4.15 KB

Installing Tensorflow with a GPU on Windows.

Prequisites (we will install all of these):

  • NVIDIA GPU on you system.
  • Anaconda with python(3.x)
  • CUDA Tool Kit
  • CuDNN

Please follow the instructions in the same order as in this text.

Installing CUDA Tool Kit:

  • Download the latest version from here:
  • After it has been downloaded open the .exe file and follow the onsceen instructions.
  • It will check for the system compatibility.
  • Then agree to the license agreement.
  • Then click on Express-> Next.(This may take some time to install and your screen may flicker at the end of the installation,but don't worry)

Adding Environment Variables.

  • This has to be done before proceeding forward and is an important step.

  • Go to Start and search for Environment Variables.

  • Click on edit environment variables.

  • Click on Path and then click on edit.

  • Then add the following paths:

    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin
    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\libnvvp
    • C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\extras\CUPTI\libx64

If the edit variables line is just a single line then add ; at the end of each of the 3 paths above and type them in a single line.

NOTE: Check the version number before adding the path.If you version is 9.2 please change it in the above paths by replacing the v9.0 with v9.2

Download CuDNN:

  • Go to this site and download the the CuDNN file for the version of CUDA you downloaded in the previous step.
  • You may have to create a profile with NVIDIA first before downloading.
  • Extract the zip file contents to someplace in the computer you wish.
  • For eg: I extracted it at C:\cudnn-9.0-windows7-x64-v7.1
  • Go to that folder->cuda->bin and copy the address(filepath) of that file.
    • for example:C:\cudnn-9.0-windows7-x64-v7.1\cuda\bin
  • Add the path to the environment variables as done before.

Update the GPU Driver:

  • First check for the GPU in your system.
  • Go to Start->NVIDIA Control Panel->System Information. In the Items list check for the GPU name. For me it is Quadro M4000.
  • Now go to this site. Fill in the details of the GPU name and download the the driver.
  • Run the .exe file.

Installing Python 3.5/3.6 in Anaconda:

  • Go to this site and download the Anaconda for python v3.5 or 3.6. *Check the add python to your PATH option during the install.

Install Tensorflow:

  • Open an Anaconda Prompt window and type the following command.

    pip install --ignore-installed --upgrade tensorflow-gpu

  • If you are under a firewall,you may need to proxy.

    pip install --ignore-installed --upgrade tensorflow-gpu --proxy <your proxy name>

Test your Installation:

  • Open another Anaconda prompt shell.
  • Type python.
  • You may want to run the prompt as an Admin if you are getting PermissionError: [WinError 5] Access is denied as an error.
import tensorflow as tf
hello=tf.constant('Hello,Tensorflow')
sess=tf.Session()
print(sess.run(hello))

  • You should get some information on the terminal about your GPU like this: Output1

  • The output should be Hello,Tensorflow

  • If you type sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)), you should get something like this:

Output2

Common Errors:

1:If you get this error:ImportError: Could not find 'cudart64_90.dll'. TensorFlow requires that this DLL be installed in a directory that is named in your %PATH% environment variable. Download and install CUDA 9.0 from this URL: https://developer.nvidia.com/cuda-toolkit Then check whether the file cudart64_90.dll is in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0\bin and you have entered the path properly in the environment variables.

2:PermissionError: [WinError 5] Access is denied: Check if you are running the prompt as an Admin