Nvidia RTX 2070 setup with Tensorflow and Keras on Windows 10

This is a guide on how to setup a deep learning environment on Windows 10 to use with Nvidia RTX 2070.

Prerequisite:

• Windows 10
• GPU Driver installed (You can download it from here https://www.nvidia.com/Download/index.aspx?lang=en-us)

Steps:

• Download and install Anaconda https://www.anaconda.com/distribution/
• Download and install Visual Studio 2017 community version RC.
• Download and install CUDA Toolkit 10 https://developer.nvidia.com/cuda-downloads.
• Download CuDNN https://developer.nvidia.com/cudnn. (You will need registration for this step)
• Extract CuDNN package and copy the below files into the CUDA directory:
• Copy the following files into the CUDA Toolkit directory
  1. Copy <cudnnpath>\cuda\bin\cudnn64_7.dll to <cudapath>\CUDA\v10.1\bin.
  2. Copy <cudnnpath>\cuda\include\cudnn.h to <cudapath>\CUDA\v10.1\include.
  3. Copy <cudnnpath>\cuda\lib\x64\cudnn.lib to <cudapath>\CUDA\v10.1\lib\x64.
• Add the environment variables CUDA_PATH, CUDA_HOME AND CUDA_PATH_V10_1. Set value to “<yourcudapath>\CUDA\v10.1″
• Add “<yourcudapath>\CUDA\v10.1\bin” and “<yourcudapath>\CUDA\v10.1\libnvvp” to your PATH variable.
• Download and install OpenSSL and add its bin folder path into environment variables.
• You may need to restart your PC for changes to take effect, then create a new environment on Anaconda with python version 3.5.2 and Tensorflow for GPU package.

  conda create --name DL tensorflow-gpu python=3.5.2
  
  activate DL

• Execute below to upgrade Tensorflow, install Keras and other Deep Learning libraries.

  pip install --ignore-installed --upgrade tensorflow
  
  update packages from ui
  
  install keras from ui
  
  conda install -n DL scikit-learn --yes
  
  conda install -n DL pillow --yes
  
  conda install -n DL matplotlib--yes
  
  conda install -n DL opencv--yes
  
  conda install -n DL numpy==1.15.4 --yes