Enable tensorflow gpu

How to setup tensorflow with GPU support on local pc

Requirements

cpu: intel i7
ram: 16G
harddisk: ssd
graphics: nvidia gtx 1060
os: windows 10

check that graphics card is supported
cuda
use gpu-z to check cuda compute version

install latest driver for graphics card
check windows device manager

NVIDIA® GPU drivers —CUDA 10.0 requires 410.x or higher.

signup with nvidia developer program

install cuda

install cudnn

Installation

1. install anaconda python

2. setup env

    ```
    conda create -n keras35 python=3.5
    conda activate keras35
    pip install tensorflow-gpu
    pip install --ignore-installed --upgrade tensorflow-gpu
   
    ```
   

3. run test program with cpu tensorflow_test.py cpu 1000
4. run test program with gpu tensorflow_test.py gpu 1000

test program available at:
tensorflow_test.py

todo enable tensorflow gpu in jupyter notebook

older version tf 1.13
conda create -n tensorflowgpu113 pip python=3.5
conda activate tensorflowgpu113
pip install tensorflow-gpu==1.13.1