Fast End-to-End Multi-Conjugate AO Simulations Using Graphical Processing Units and the MAOS Simulation Code
TMT Observatory Corportaion, 1111 South Arroyo Pkwy Suite 200, Pasadena, CA, 91105
AO for ELT conference, 2012
@article{wang1a2012fast,
title={Fast End-to-End Multi-Conjugate AO Simulations Using Graphical Processing Units and the MAOS Simulation Code},
author={Wang1a, L. and Ellerbroek, B.},
year={2012}
}
The Multi-threaded Adaptive Optics Simulator (MAOS) was developed at TMT to efficiently simulate various kind of AO systems. In particular, it can finish a time step of full end-to-end simulation of an ELT size multi-conjugate AO system in 1 second on 8 contemporary cpu cores. We recently ported it to run on graphical processing units (GPUs) using the Nvidia CUDA technology. A 10 fold speed up is obtained with two GTX 580 GPUs, with each time step taking only 0.1 second. A single GPU can finish 30 iterations of the conjugate gradients (CG) tomography algorithm in 30 ms, or 3 iterations of the fourier domain preconditional CG in 5 ms, which is on the same order as the ~1ms requirement.
July 25, 2012 by hgpu
