Programming strategy for pascal/TX2 memory hierarchy

I am new to nvidia and I’m getting started on programming a pascal GPU plus a TX2. As I understand it, the Pascal architecture introduces a higher level of memory sharing between CPU and GPU. Can anyone point me to good information on:

How does unified memory affect CUDA strategy?
Can one be too reliant on the unified memory architecture?
What are the tradeoffs for unified memory?

Thanks, —Marc.