The continuously increasing demands in terms of performance, environmental compatibility and safety motivate the growing interest for optimal control in automotive systems. In practice, however, these methods are seldom used, one of the reasons being the nonlinear nature of the plant which makes the computation more difficult. Several nonlinear optimal control methods have been tested for automotive systems, either starting from the optimization problem in terms of a Hamilton Jacobi Bellman equation or as a nonlinear extension of well established model predictive approaches. In this paper, we apply a method which in some sense combines both worlds, the C/GMRES method. In our implementation, no physical model is used but a nonlinear NARX model is derived from data. The resulting control law is applied to a production Diesel engine and tested against the production controller, showing the potential performance of the suggested method but also the design simplicity.