Performance assessment of control loops has got the attention of many researchers. This study focuses on performance assessment of model predictive control. An MPC-achievable benchmark is proposed based on subspace identification; the advantage of this benchmark is that only system input and output data is used and the state space model need not explicitly calculated during identification, in addition, it can be implemented to both univariate and multivariate process. Two performance measures can be constructed to assess the control loop, which can evaluate the potential benefit to update the new identified model. To find the potential benefit by tuning the parameter, tradeoff curves similar to LQG benchmark can be draw as reference. The effect of constraints imposed on process variables can be evaluated by the installed controller benchmark. Simulation about an industrial example was done using the proposed method.