Title
Numerical Solution of the Wave Equation on Dual-GPU Platforms Using Brook+
Document Type
Presentation
Publication Date
4-12-2010
Faculty Sponsor
Dr. Inanc Senocak
Abstract
The continuing motivation for our research is to accelerate the simulation of contaminant dispersion in urban environments. Computational turn-around time for every stage of the simulation task needs to be fast because of emergency response requirements. Additionally, we investigate the stream computing paradigm to accelerate numerical computations of partial differential equations on the modern graphics hardware. In particular, we assess the computational performance of the Brook+ programming language on AMD's FireStream boards with 320 streaming cores relative to a conventional processor. Brook+ is a high level language that handles all the memory copying and allocation on both the central processing unit (CPU) and graphics processing unit (GPU) Further optimization is made by running a streaming program in a parallel manner on multiple FireStream boards. A second level of parallelism adds new levels of complexity and over head. We look to compare the performance of a multi-GPU implementation compares to the single GPU and serial CPU program. As well as compared to parallel CPU implantation of the serial program.