When time step is too small and time lapse is 20ms , it might exceed buffer or cache limits or something in LTspice. Use 2ms or 5ms stop time which works.
I've sometimes found that tinkering with some other component than the one LTS points to effects a cure. Capacitors and inductors sometimes benefit from an internal parasitic parameter being set.
This also happens when the simulation creates derivatives with very steep slopes. As it tries to make the time step small enough to provide for such a slope it reaches a limit. There are no simple or easy - works every time solutions. The process of elimination is about the only thing that works.
Your circuit lacks phase compensation which due to lack of phase margin causes oscillations on start-up, which is easily fixed, but the no-load noise seems excessive for the values chosen after the output reaches target.
My simulation added a 0.2 ohm load.