FMM is an efficient algorithm in computing N-body problem. This paper firstly subdivides the FMM into 10 procedures. Based on the analysis the computing type of each procedure, we choose key procedures accelerated on FPGA, GPU and Cell BE. And then we present the speedup ratio of each accelerated procedure through experiments. Finally we analyze the computing characteristic of FMM on the computing architecture on accelerator FPGA and GPU on the side of P, M and C.