high performance computing on graphics processing units: hgpu.org

We develop a hybrid multiGPUs and CPUs version of an algorithm to model seismic wave propagation based on the spectral-element method in the case of models of the Earth containing both fluid and solid layers. Thanks to the overlapping of communications between processing nodes on the computer with calculation by means of non-blocking message passing, we obtain excellent weak scalability of this finite-element code on a cluster of 192 GPUs and speedup factors of more than one order of magnitude compared to the same code run on a cluster of traditional CPUs. This enables us to show a new geophysical phenomenon concerning wave propagation of diffracted shear waves in a layer called D” located at the base of the Earth’s mantle, namely that in this layer the transverse and radial components of these waves can undergo a relative shift even in an isotropic Earth model, whereas this observation in real seismological data was interpreted until now as an indication of the presence of anisotropy in this layer.