This paper presents a model predictive online optimization scheme for the engine torque control problem. The control-oriented model is based on the intake air charging dynamics and torque generation model which are derived from the mean value model. In order to reduce the tracking error induce by the insufficient accurate predictive model, an embedded integrator about the tracking error is designed. Then, the online optimization algorithm namely C/GMRES is adopted to solve the nonlinear optimal problem. Finally, experimental validations are conducted and demonstrated the robustness and transient performance for the proposed engine torque controller.