It was all working fine (Simply install keras-gpu under Anaconda) , until I upgraded my Anaconda Environment ....
Now I have to reinstall Tensorflow and Keras into my Windows Anaconda distribution. But Tensorflow 1.5 is dependent upon Cud 9.0, so I have had to go back an install Cuda 9.0, and CudaNN 7.0. But I am now pretty sure that the Anaconda Distribution is still out of whack, as Anaconda only references Tf at v1.2.1 and Keras at version 2.1.3.
Can Keras 2.1.3 make use of Tensorflow 1.5 ?
Disappointing that Anaconda is supposed to provide a consistent ML distribution environment, but doesn't maintain Tensorflow or Keras in step.