A parallel code named P-EFIT which could complete an equilibrium reconstruction iteration in 220 µs is described. It is build with the CUDA™ architecture. Some optimization for middle-scale matrix multiplication on graphics processing unit and an algorithm which could solve block tri-diagonal linear system efficiently in parallel is described. Benchmark test is conducted. Static test proves the correctness of the P-EFIT and simulation-test proves the feasibility of using P-EFIT for real-time reconstruction with a 65 × 65 grid.