Unless you write ir explicted in the script, it will use a slightly less tolerance for the equivalent bulk run, as you can save quite sometimes since a equivalent bulk converged to the 1e-3 (or 1e-5) gives the more or less the same SCF cycle for the device part of the calculation.