"Closed-loop" seems to have developed a squishy definition. Going from least effective to most effective:
1) Encoder monitoring an open-loop step motor. The monitor gives an error output if steps taken don't match steps sent.
2) Encoder terminal correction. An open-loop step motor is commanded to move from A to B. If the number of steps taken don't match the number sent, the correction circuit sends steps equal to the difference at the end of the move to null the error.
3) Encoder following error correction. The motor operates open-loop. If it's following error approaches +/-1.8 degrees, the drive goes closed-loop until the following error is reduced to below the limits.
4) Tightly coupled closed-loop. The loop position error output (PID filtered) makes continuous motor corrections on a step-by-step basis while the motor is running or stopped. This is what most people think of as being closed-loop. DC servomotors cannot be run any other way.
5) Adaptive closed-loop. Just like (4) except the loop circuitry also generates a feedback command to the motion controller telling it to slow down if the motor begins to develop a continuously increasing following error. The controller slows down until the motor can "reel-in" the following error.
Let's stick with definition (4). Loop bandwidth determines how quickly the motor can respond in correcting an error. The G320 and G320X use analog circuits that have a 2.5kHz bandwidth. If it were done digitally, the equivalent sampling rate (number of corrections per second) has to be 5,000 per second (200uS update period).
Mach3 is not designed to operate closed-loop, meaning Mach3 cannot read the encoder feedback, compare it against a commanded position, PID filter the difference and output step pulses to null the difference to zero. That is the reason why the loop is closed locally in the G320 and G320X. The result is the same as in a distributed servo loop (PID controller in the PC and a torque command signal to a remote motor amplifier) except that fault or error signals back to the PC are harder to implement.
The performance of such a distributed system depends largely on the loop bandwidth. A 100Hz bandwidth system will be far more sluggish than a 3kHz system.
Mariss