Installing tensorflow-gpu on windows is same as tensorflow (cpu only). You can use the below pip command in winpython command prompt.
pip install tensorflow-gpu
This will install tensorflow-gpu on windows, but to ensure tensorflow-gpu works on your graphic device, you need to make sure to install necessary compatible drivers of your graphic card.
Using NVIDIA Graphics Card for Tensorflow-gpu
To use the NVIDIA graphics card as GPU for your tensorflow you need to install the following.
Step 1: Install CUDA Toolkit : (https://developer.nvidia.com/cuda-downloads)
Step 2: Install CUDNN (https://developer.nvidia.com/cudnn) (You need to sign up for nvidia developer to download this)
Step 3: Extract the CUDNN from step 2 and copy the folders (bin, include, lib) to the cuda folder (C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0). This will add cudnn library to CUDA.
Now test the tensorflow-gpu by typing in the python console
If you still get the errors. You have incompatibility with versions of python, tensorflow, cuda, cudnn.
The incompatibility with these versions is the common question in stackoverflow which keeps popping the same question again for every new version.
The latest version which works is –> Python 3.6.2, Tensorflow1.3.1, CUDA 8, CUDNN 6 on windows 10