Achieving high solution efficiency for alternating current power flow (ACPF) analysis from high-performance computing (HPC) architecture is a leading and important challenge in power system analytics and computation. This paper investigates the performance of the fast decoupled (FD) method, which is based on the direct linear solver and implemented on the graphics processing unit (GPU), for the solution of ACPF. Implementation platforms, linear equations solution strategies, data storage formats, and fill-in reduction algorithms are compared and discussed on five benchmark systems ranging from 300 to 13,659 buses. Within the GPU’s compute unified device architecture (CUDA) environment, the shortest ACPF solution time for the largest test case is 0.313s, which is 4.16× faster than its Matlab counterpart.