A new time optimal model predictive control paradigm is proposed for the linear time-varying system with bounded state disturbance. The suboptimal polyhedral N-step reachable sets are determined offline by solving a series of linear programs, and then the inputs are optimized online to render the states into the terminal set as fast as possible. This method can handle asymmetric constraints. Compared with previous methods, it avoids the possibility that the number of vertices increases exponentially with the step N, and thus eliminates excessive complex polytope operations. The results show that the proposed approach is convenient for practical purposes.