How important is the size of the k-mesh in the nscf file?

[CHILLL]

Dear experts and users,

I have noticed that in the majority of articles where EPW was used people talk about k-mesh of the self-consistent calculation and q-mesh of the phonon calculation. For example, in the PRB article "Anisotropic Migdal-Eliashberg theory using Wannier functions" in the case of MgB2 different k-meshes and q-meshes were considered, but the size of the mesh for the nscf calculation is not really discussed much.
So, I have been wondering if the nscf kgrid should be of the same size as scf, or it can be much smaller? I understand that one should check the convergence, but maybe there is a reasonable minimum.

Best, Mikhail