How to force devices to run in sequence for debugging

Accelerated Computing CUDA CUDA Programming and Performance
#1

How can it be done?