Two GPUs Required

I am getting started with CUDA and the documentation says I need two CUDA enabled GPUs for local debugging. So my questions are:

  1. Do really both have to support CUDA or only one of them?
  2. Do they or should they be the same or can I use a low-end version for display and a high-end card as the target?
  3. Could I also get away with just one CUDA enabled GPU? If so what are the disadvantages?