Abstract: Currently, most of the commercial finite element (FE) softwares are based on the CPU architectures, which causes massively time consuming, low efficiency, and rigor of requirements of hardware during analyzing the nonlinear response of high-rise structures. Meanwhile, the emergence of GPU based algorithms presents significantly superior performance in floating-point operation and parallel computation due to its special configuration. Therefore, GPU based algorithms can provide a feasible solution for the perplexing issues of nonlinear computation of high-rise structures. Our work is to develop a parallel FE algorithm by introducing GPU and to construct a corresponding heterogeneous platform, ultimately leading to speed up the computation. Firstly, the mapping between the degrees of freedom (DOFs) of a refined model and the threads of GPU is formed. Then, the implicit integration algorithm for solving the dynamic response will be parallelized in threads; meanwhile, the strategies of storage are optimized in terms of element-by-element scale and the demand of memory was reduced while solving the equations. All of the GPU based algorithms have been validated by comparing with the experimental results of a shaking table. Moreover, the validated algorithms are extended to apply to the analysis of the elastic-plastic seismic response of a practical high-rise reinforced concrete frame tube structure. The results show that the proposed algorithm can not only guarantee the accuracy but also improve efficiency dramatically in the procedure of structural nonlinear response analyses.