## The solution of the Marmousi problem on GPU's

During the PhD research of Hans Knibbe one of the first results (July 2011) obtained on the
LGM
is the solution of the Marmousi problem.

The Helmholtz equation was used to compute the wave propagation in a medium
of constant density but variable velocity. About 100 frequencies were
computed in parallel using MPI on 15 GPUs (1 GPU per node used) of the LGM.
The computation took 30 minutes We used first order radiation boundary
conditions.

The Marmousi model can be found at the following
site.

The source is a 15Hz Ricker wavelet. The computational grid is the same as
the model grid (2301 x 751 grid points) as the grid is fine enough.

Click on the figure below to see the movement of the
seismic front in the Marmousi problem.

Layout of the Marmousi problem

The second movie
has been produced using the central part of the Marmousi
model. The horizontal extents are from 3068 meters to 6136. Like the
original model, the vertical extents are from 0 to 3004 meters. The model
was resampled so that dx=dy=4/3 meters.

The source was a 15 Hz Ricker wavelet placed in the middle at a depth of
2*4/3=2.6 meters.

Part of the funding for the LGM has been obtained by NWO under the project number
612.071.305.

For background information of the method used in this work we refer to:

H. Knibbe and C.W. Oosterlee and C. Vuik
(pdf,
bibtex)

GPU implementation of a Helmholtz Krylov solver preconditioned by a shifted Laplace multigrid method

Journal of Computational and Applied Mathematics, 236, pp. 281-293, 2011

A
presentation
has been given at the ENUMATH Conference 2011.

## Contact information:
Kees
Vuik

Back to the
home
page
of Kees Vuik