Accelerate Double-Difference Seismic Tomography with GPU Architecture
  Pei-Cheng Liao     Yu-Chi Lai     Ping-Yu Chang     Sheng-Rong Song  
Pei-Chang Liao, Yu-Chi Lai, Ping-Yu Chang, Seng-Rong Song, Accelerate Double-Difference Seismic Tomography with GPU Architecture, 2012 AGU Meeting

Double-difference (DD) seismic tomography uses both absolute arrival times and differential travel times at different observation points to estimate the velocities of seismic P- and S-wave in the measurement area to estimate the locations of the earthquakes and the underground velocity structures. Generally the number of observation locations involved is large in such inversion task, and the amount of computation involved in estimating the velocity of waves is huge. The facts imply that the computation time required to calculate the earthquake location and recover the underground structure of the interesting area is large. In the other hand, one could find that the estimation and reconstruction processes contain a large number of independent matrix computations and thus, the process should be easily accelerated by parallel processing techniques. Conventionally the parallel processing computing is executed with costly cluster computers or the supercomputer. Recently the low-cost Graphic Processing Units shred the light for cheaper parallel computing option for those who cannot afford the expensive computer systems. In the study, we use the Compute Unified Device Architecture (CUDA) to implement the double-difference seismic tomography algorithm to explore the possibility of accelerating the computation with a parallel processing many-core GPU. The experimental result shows that the GPU-based DD algorithm is much more efficient than CPU-based DD algorithm in the example cases. 


author = {Pei-Chang Liao and Yu-Chi Lai and Ping-Yu Chang and Seng-Rong Song},
title = {{Accelerate Double-difference Seismic Tomography with GPU Architecture}},
booktitle = {AGU Meeting},
year = {2012}