|The geometric multigrid method (GMG) is one of the most efficient solution techniques for
discrete algebraic systems arising in many types of partial differential equations. GMG utilizes
a hierarchy of grids and discretizations and reduces error of different frequencies simultaneously.
Graphics processing units (GPUs) have recently burst onto the scientific computing scene as a
technology that has demonstrated substantial performance and energy efficiency improvements.
However GPU is not particularly efficient to coarse level problems in GMG. In this work, we
study the performance of GMG on CPU-GPU heterogenous computers. Furthermore, for comparison,
we compare our method with an efficient implementation on CPUs as well as the Fast
Fourier Transform in CuFFT.|