APPLICATION OF CUDA TECHNIQUE TO QUICKEN CALCULATION OF EUROPEAN OPTION PRICES BY FINITE-DIFFERENCE METHOD
ITMO University; Department of Computer Technology;
D. N. Shinkaruk
ITMO University; Post-Graduate Student
A. V. Toropov
Saint Petersburg National Research University of Information Technologies, Mechanics and Optics; Tbricks AB; postgraduate, engineer-programmer
Y. A. Shpolyanskiy
ITMO University; Professor
Abstract. Crank—Nicolson scheme for Black–Scholes partial differential equation is fully implemented on graphics processor using CUDA technique. The developed code on GPU NVIDIA GTX 580 works more than 20 times faster than the single-threaded calculation on CPU Intel Core i7 3.4 GHz, and 2—3 times faster than the best results obtained with multi-thread version based on GCD technique on CPUs 2 x Intel Xeon 3.06 GHz with 24 cores in conditions typical for high-frequency algorithmic trading systems.
Keywords: CUDA, GPGPU, parallel cyclic reduction, PCR, algorithmic trading, high-frequency trading, option, Crank—Nicolson scheme.