high performance computing on graphics processing units: hgpu.org

The Finite-Difference Time-Domain (FDTD) method is a computational technique for modelling the behaviour of electromagnetic waves in three-dimensional space. When executed to solve real-world problems the FDTD method is characterised by long execution times involving a large amount of data organised into matrices. The FDTD method exhibits ample data parallelism, and parallel computing techniques are frequently used to reduce the execution time of FDTD computations. This project explores the opportunities for accelerating the computation of the FDTD method using two types of commonly available parallel hardware. First, the Streaming SIMD Extensions (SSE) capability of the x86 processor architecture is considered. The SSE capability of processors based on this architecture allow a single instruction to be applied to multiple sets of data simultaneously. This project investigates the possibility of using SSE to accelerate the FDTD method by performing calculations for multiple matrix elements with a single instruction. Second, general-purpose computing on graphics processing units (GPUs) is considered. As the hardware used in GPUs has become less specialised, they have come to resemble general-purpose, massively parallel processors. Typical modern GPUs have hundreds of processor cores which operate concurrently, offering the potential to greatly accelerate computations with sufficient parallelism. This project investigates the possibility of accelerating the FDTD method using GPU computing. Both hardware technologies are evaluated for the speedup they offer to the FDTD method, as well as the complexity of their implementation.