Abstract. Ice sheet models differ in their numerical treatment of dynamical processes. Simulations of marine-based ice are sensitive to the choice of Stokes flow approximation and basal friction law, and to the treatment of stresses and melt rates near the grounding line. We study the effects of these numerical choices on marine ice-sheet dynamics in the Community Ice SheetModel (CISM). In the framework of the Marine Ice Sheet Model Intercomparison Project 3d (MISMIP3d), we show that a depth-integrated, higher-order solver gives results similar to a 3D (Blatter-Pattyn) solver. We confirm that using a grounding-line parameterization to approximate stresses in the grounding zone leads to accurate representation of ice sheet flow with a resolution of ∼2 km, as opposed to ∼0.5 km without the parameterization. In the MISMIP+ experimental framework, we compare different treatments of sub-shelf melting near the grounding line. In contrast to recent studies arguing that melting should not be applied in partly grounded cells, it is usually beneficial in CISM simulations to apply some melting in these cells. This suggests that the optimal treatment of melting near the grounding line can depend on ice-sheet geometry, forcing, or model numerics. In both experimental frameworks, ice flow is sensitive to the choice of basal friction law. To study this sensitivity, we evaluate friction laws that vary the connectivity between the basal hydrological system and the ocean near the grounding line. CISM yields accurate results in steady-state and perturbation experiments at a resolution of ∼2 km (arguably 4 km) when the connectivity is low or moderate, and ∼1 km (arguably 2 km) when the connectivity is strong.