general help getting started

So I can write a Deep ANN in c (not parallel) and I can write some simple cuda but I want to write some code to train a deep network on GPU using cuda/cudnn.

So the issue is I am not sure where to start.

The mnist sample code is for inference but I want to do backprop, does cudnn hep with this at all ? Or is cudnn only doing convolutions and pooling.

If anyone can suggest any articles or anything that would be great


Please refer to RNN and Conv samples: