RTM using effective boundary saving: A staggered grid GPU implementation

Conclusion and discussion

In this paper, we introduce the effective boundary saving strategy for GPU-based RTM imaging. Compared with the method of Dussaud et al. (2008), the saving amount of effective boundary with regular grid finite difference scheme is slightly reduced. The RTM storage of effective boundary saving for staggered finite difference is first explored, and then implemented with CPML boundary condition. We demonstrate the validity of effective boundary saving strategy by numerical test and imaging of benchmark models.

The focus of this paper is RTM implementation using effective boundary saving in staggered grid instead of GPU acceleration. A limitation of this work is that the numerical examples are generated with NVS5400M GPU on a laptop (compute capability 2.1, GDDR3). It is easy to do performance analysis for different dataset size and higher stencil orders if the latest GPU card and CUDA driver are available. It is also possible to obtain improved speedup by incorporating MPI with GPU programming using advanced clusters with larger GDDR memory (Suh et al., 2010; Komatitsch et al., 2010a) or FPGA optimization (Fu and Clapp, 2011; Medeiros et al., 2011). Unfortunately, higher stencil orders of staggered grid RTM using effective boundary implementation in 3D is still a problem. 3D RTM using the 2nd order regular grid finite difference with Clayton and Enquist boundary condition (only 1 layer on each side to save) needs tens of GBs (Liu et al., 2013b). It implies that 3D RTM with higher stencil orders will definitely exceed the memory bound of current and next generation GPUs. For GPU implementation of 3D RTM, the practical way is using the random boundary condition (Liu et al., 2013a) or saving on the disk. A deeper discussion of the practical issues for GPU implementation of RTM can be found in Liu et al. (2012a).

RTM using effective boundary saving: A staggered grid GPU implementation