Firstly, I would prefer that you just call these SG-2 and SG-3, those are their names; “fine” and “ultrafine” is Gaussian terminology and I don’t guarantee that they are identical because Gaussian never published the details of their grids.
What functional are you using? The behavior with respect to grid is very functional-dependent. The default grids in Q-Chem were chosen with this in mind, with the caveat that the Minnesota functionals have integrands that are extremely oscillatory and even SG-3 (which is the default for Minnesota functionals) is occasionally not quite converged in those cases, in which case I recommend unpruned (99,590), i.e., XC_GRID = 000099000590.
As explained in our paper (https://doi.org/10.1002/jcc.24761), we wanted “standard grids” and to have something that worked well for Minnesota functionals in all cases would have required us to be too conservative for other functionals, and we don’t like the Minnesota functionals.