Full text loading...
-
Seismic Modeling with Half Precision Floating Point Numbers
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, 80th EAGE Conference and Exhibition 2018, Jun 2018, Volume 2018, p.1 - 5
Abstract
Half precision floating point numbers is becoming increasingly supported by new processors, often with a significant throughput gain over single precision operations. In this article, we investigate whether half precision is suitable for finite difference based seismic modeling in the context of imaging and inversion. By scaling the finite difference expression of the isotropic elastic wave equation, we manage to obtain a stable solution despite the very narrow dynamic range of the half-precision format. We present a CUDA implementation of this code, which, on most recent GPUs, is nearly twice as fast and uses half the memory of the equivalent single precision version. The error on seismograms caused by the reduced precision is shown to correspond to a fraction of a percent of the total seismic energy, and is mostly incoherent with seismic phases. Thus, half precision modeling could accelerate full waveform inversion or migration with a negligible impact on the quality of their output.