On some algo (e.g. PHI1612), enemy initialize my cards very slowly ( 10+GPU rigs on W10, 390+ ), I mean minutes.
I have experienced this "issue" with ccminer too, but a "--cuda-schedule 12" command solve my problem without high CPU usage (if I use any other number as 12 my CPU usages go to the moon and the system becomes unmanageable).
In enemy, I tried all options:

--cuda-schedule     set CUDA scheduling option:
  0: BlockingSync (default)
  1: Spin
  2: Yield

but unfortunatelly did not help (max CPU usage, except with 0).

Can you add the "old / original" --cuda-schedule options?