Add lots of checks to CUDA calls in the timer to try to identify any
problems that might be showing up on the dashboard.

Also adding some print statements around the sleep function in the
device adapter testing. For some reason the problem just went away with
them.