SD in vacuum and domain decomposition fatal error

GROMACS version: 2024.5
GROMACS modification: No

I am running stochastic dynamics simulations of many molecules in vacuum at low concentration (~1 microM). The box is almost empty so collisions among molecules are infrequent. I am using a relatively long tau_t (25 to 100 ps) and when using domain decomposition I often get this fatal error:

One or more atoms moved too far between two domain decomposition steps.

this does not happen using shorter tau_T (1 ps) where molecules do not move much around, or when using more concentrated systems where collision are more frequent.

My question is, is that a true fatal error in the sense that the DD algorithm cannot work correctly under those conditions or it is a safeguard error because if this happen in a dense system it clearly indicates something very far from equilibrium?

The issue is solved when compiling in double precision, suggesting that for long tau there is an error accumulation issue in the integration of the equation of motion.