基于CPU-GPU异构并行的复杂场地近断层地震动谱元法模拟

Simulation of near-fault ground motions in complex sites based on CPU-GPU heterogeneous parallelism by spectral element method

  • 摘要: 利用基于CUDA编程平台实现的工作站级CPU-GPU异构并行方法开展了实际场地近断层地震动谱元法模拟。通过模拟SECE/USGS提供的自发破裂模型TPV15,测试了工作站级CPU-GPU异构并行方法的计算精度与计算效率,并将该方法应用于1679年三河—平谷M8.0地震的强地面运动模拟,以证实该方法对真实设定地震动模拟的适用性。模拟结果显示:CPU-GPU异构并行计算时间较CPU并行计算时间明显减少,加速比最高值分别为CPU 36核和72核的3.04和2.16倍;1679年三河—平谷M8.0地震的强地面模拟结果清晰地体现出近断层地震动的集中性、破裂的方向性、速度脉冲和永久位移等近断层地震动特征以及真实地形对近断层地震动所产生的影响。结果表明,CPU-GPU异构并行方法有效地提高了谱元法模拟的计算效率,可应用于大尺度复杂场地地震波场模拟。

     

    Abstract: Base on CUDA programming platform, the workstation-level CPU-GPU heterogeneous parallel method is implemented, and the spectral element method is used to simulate ground motion near-fault in a real site. In this paper, the computational accuracy and efficiency of the proposed workstation-level CPU-GPU heterogeneous parallelism method are tested by simulating the spontaneous rupture model TPV15 provided by SECE/USGS. Furthermore, the proposed method is applied to the simulation of strong ground motion in 1679 M8.0 Sanhe-Pinggu earthquake, and therefore the applicability of the proposed method to the simulation of real ground motion is verified. The simulation results show that the computing time of CPU-GPU heterogeneous parallelism is significantly reduced than that of CPU parallelism, and the highest acceleration ratio is 3.04 and 2.16 times as long as CPU 36 core and 72 core respectively. The simulation results of M8.0 in Sanhe-Pinggu earthquake in 1679 clearly show the characteristics of near-fault ground motion, such as near-fault ground motion concentration, fault rupture directivity effect, velocity pulse and permanent displacement, and the influence of real terrain on near-fault ground motion. The results show that the CPU-GPU heterogeneous parallelism method can effectively improve the computational efficiency of spectral element method simulation, and it has a good prospect to be applied to seismic wave field simulation of large-scale complex sites.

     

/

返回文章
返回