Integrating Multiple Input Multiple Output (MIMO) into Multi-access Edge Computing (MEC) as a new computing paradigm can provide users with higher quality of services. In this paper, a cloud-edge-end three-layer collaborative MIMO-MEC system is considered, in which multiple MEC servers can cooperate with each other. We describe the problem as a joint optimization of offloading strategy, computational resource allocation and downlink power allocation to minimize the total delay of all users. A nested algorithm is proposed to address the mixed-integer non-linear programming problem. Therein, a penalty function-based Newton method and the Lagrange multiplier method are used in the inner layer, while a heuristic algorithm is adopted in the outer layer. The results show that the proposed scheme can effectively reduce the total delay of processing tasks compared with the baseline schemes.