Section: Application Domains

Financial computation

AmerMal application is a parallel American Option Pricer developed in collaboration with Lokman Abbas-Turki during his PhD thesis at University Paris-Est. This pricing mathematic algorithm has been designed in order to be parallelized on clusters of GPUs. It is based on a Maillavin calculus and on Monte Carlo stochastic computations. However, the Monte Carlo trajectories are coupled, and our final parallel algorithm includes many communications between the computing nodes. Nevertheless, we achieved good scalability on our 16-nodes GPU clusters.

The initial version of AmerMal has been implemented using MPI+OpenMP+CUDA in 2010 and 2011. In 2012, we have ported this application on ORWL+CUDA, both in order to test and improve ORWL semantic and ORWL performances.