Another possible explanation is the drv8825 blanking time (missing microsteps in the low part of the voltage range).
Take a look at the excellent analysis here: [cabristor.blogspot.ca]
The short form: Put a pair of back to back diiodes in-line with each coil of each stepper motor. The original author recommends 4 diodes for each coil, but in my tests a single pair seemed to do the job.