The traditional Security Constrained Optimal Power Flow (SCOPF) methods are limited for the realistic systems due to the high memory consumption and low computation efficiency. A high-speed multi-core parallel algorithm is proposed to overcome these disadvantages. The contingencies are checked using the pre-contingency configuration at first. Afterward, a few serious contingencies are extracted to correct gradually the normal operation status until all conditions are satisfied. The multi-core parallel technology is applied to the algorithm from both high and low levels for obtaining high efficient parallel computation. The high-level parallels are used during the contingencies checking procedure, and fine-grained parallels are used during the pre-contingency configuration correction process based on the directed acyclic graph (DAG) parallel numerical decomposition algorithm. The proposed method has been successfully tested on three systems, which include a realistic provincial 3301-bus system with 693 contingencies. The numerical results indicate that the proposed method is suitable for solving large SCOPF problems because of fast and steady computation time. Furthermore, the memory consumptions of the proposed method are less 1% than that of traditional method, and the computation speeds are three orders of magnitude faster than that of the traditional ones as the system scale enlarges.