engleski

EUROPEAN SWAPTION BY MONTE CARLO ESTIMATION ON MODERN INTEL ARCHITECTURES

Modern computing resources provide substantial processing power but require that the programmatic implementation of used numerical algorithms is fine-tuned to the target CPU architecture. The main performance gains could be obtained minimizing cache traffic and by efficient using of multiple levels of CPU parallel units. Monte Carlo estimation of LIBOR models is an example of challenging computational finance/computer programming problem. This paper compares the efficiency of multicore (Intel Core Haswell) and manycore platforms (Intel Xeon Phi Knight Corner and Intel Xeon Phi Knight Landing) for the calculation European LIBOR-based swaptions. Target optimized programmatic implementations of LIBOR calculation using the Intel Cilk Plus and OpenMP standards are presented and benchmarked. Results show that Intel Xeon Phi Knight Landing evaluate payoff European LIBOR-based swaptions faster, economically profitable and more energy efficient than the Intel Xeon Phi Knight Corner coprocessor and Intel Core Haswell.

Libor ; Monte Carlo simulation ; Intel Cilk Plus ; OpenMP ; Intel Knights Landing

nije evidentirano