This work develops and compares three different controllers for PEMFC systems in automotive applications. All the controllers have a cascade control structure, where a generator of setpoints sends references to the subsystems controllers with the objective to maximize operational efficiency. To develop the setpoints generators, two techniques are evaluated: off-line optimization and Model Predictive Control (MPC). With the first technique, the optimal setpoints are given by a map, obtained off-line, of the optimal steady state conditions and corresponding setpoints. With the second technique, the setpoints time profiles that maximize the efficiency in an incoming time horizon are continuously computed. The proposed MPC architecture divides the fast and slow dynamics in order to reduce the computational cost. Two different MPC solutions have been implemented to deal with this fast/slow dynamics separation. After the integration of the setpoints generators with the subsystems controllers, the different control systems are tested and compared using a dynamic detailed model of the automotive system in the INN-BALANCE project running under the New European Driving Cycle.