In the M880A manual, it says that the minimum step pulse length is 1.5us and on direction change, the direction signal should change at least 5us before the next step pulse. There are two pieces of information that are very important here that you have not given us. First of all, what motion control software are you using? Typically, this is going to be Mach3 or LinuxCNC but you might be using something else. Secondly, how does the motion control software connect to the stepper drivers? This might be parallel port to breakout board (BOB), or USB or Ethernet to a motion control board. The signal timings are generated by Mach3 or LCNC if you are using a parallel port, and by the motion control board if you are using one. Without knowing what these are, we cannot say much more at this stage. Typically, if there is a pulse timing issue, it is because the motion controller does not wait long enough after changing the direction signal before it starts sending step pulses. However, the M880A is pretty fast according to the manual, so I would be surprised if the default settings for whatever you are using are not OK. However, as I say, without knowing what you are using and if it has been changed from the default, this is all just guesswork.

If the step pulse polarity is wrong, then generally you are going to see a drift of position in one direction only. It is not easy to test this directly - you can measure the "drift" but that doesn't tell you where it is coming from. One way to check is to change the step pulse polarity (which is fairly easy to do in the "Ports and Pins"->"Motor Outputs" page in Mach3, and the equivalent in LCNC but I can't now remember exactly how you get there) and see if the problem goes away. Alternatively, what I found on my own machine when I had this problem was that I lost the equivalent of one step pulse every time there was a direction change from + to -. On my machine, one step pulse is about 0.003mm. By using a dial gauge to measure the position of the axis, I could run gcode that just went backwards and forwards, and I found that the position error after X direction changes + to - was X times 0.003mm. In other words, I could actually see that the error increased by one step per direction change, whatever number of direction changes were made. Because a single step error is very difficult to measure, you do need to do, say, 100 direction changes to get a measurable error. I must say that I did check as many possible mechanical areas as I could, particularly where something could slip a tiny amount - things like motor shaft couplings, for example - before I tracked this down to a wiring fault (I had a twist in my cable, in effect, but I could cancel this out via the ports and pins setting as above).