Section: Application Domains

Scientific computation

In the context of studying GPU cluster programming as well as asynchronous algorithms, we have developed a parallel code to solve a PDE problem that is representative of scientific computations. This is a 3D version of the advection-diffusion-reaction problem modeling the interactions of two chemical species in shallow waters. Different versions of the application have been implemented. A first version has been the subject of extensive studies about computing and energy performance. Other versions, especially the last one using MPI+OpenMP+CUDA, are still the subject of studies.

Also in a close collaboration with Fatmir Asllanaj from the LEMTA laboratory, Nancy, France, we launched the development of a code for Radiative Transport Equations, ETR. As a first step to higher efficiency the code (originally in FORTRAN) was completely rewritten in C. Already this sequential rewrite has largely improved the time and memory efficiency of the code. The next step will be a parallelization for multi-core machines and clusters that will make this a unique tool to tackle ETRs in 2 and 3 dimensions.