Efficient pseudo-random number generation for monte-carlo simulations using graphic processors
CERN, CH-1211, Geneve 23, Switzerland
Journal of Physics: Conference Series, Volume 368, 2012
@article{1742-6596-368-1-012024,
author={Siddhant Mohanty and A K Mohanty and F Carminati},
title={Efficient pseudo-random number generation for monte-carlo simulations using graphic processors},
journal={Journal of Physics: Conference Series},
volume={368},
number={1},
pages={012024},
url={http://stacks.iop.org/1742-6596/368/i=1/a=012024},
year={2012}
}
A hybrid approach based on the combination of three Tausworthe generators and one linear congruential generator for pseudo random number generation for GPU programing as suggested in NVIDIA-CUDA library has been used for MONTE-CARLO sampling. On each GPU thread, a random seed is generated on fly in a simple way using the quick and dirty algorithm where mod operation is not performed explicitly due to unsigned integer overflow. Using this hybrid generator, multivariate correlated sampling based on alias technique has been carried out using both CUDA and OpenCL languages.
February 17, 2014 by hgpu
