FiNS: A framework for accelerating nested simulations on heterogeneous platforms

Abstract

Insurers use advanced risk management models to, among other things, compute required capital for different sources of financial risks. In these models the application of nested simulations becomes increasingly important. To keep computation times within acceptable limits high-performance computing is required. In this work we present a framework designed to significantly improve the performance of nested simulations by using heterogeneous computing. Specifically, we use modern features from CUDA - streams, Hyper-Q, and Multi-Process Service - to take full advantage of the massive parallelism of modern GPUs. We manage to reduce the execution time of such simulations from several hours to tens of minutes.