Skip to Main Content
Modern graphics processing units (GPU's) utilize a programmable parallel pipeline architecture to render complex scenes onto a two-dimensional screen. Rendering these scenes requires rasterization, texturing operations, and multiple stages of lighting operations. These processes are computationally intensive and must be performed near real-time in today's gaming and workstation applications. These industries have driven the performance of GPU's to exceed that of CPU's in terms of operations per second. Much effort has been placed recently on harnessing the power of the GPU for general purpose computation. In this paper, Accelerator by Microsoft Research provides an interface to the GPU using a library of classes and functions in Microsoft C Sharp (C#). The performance of a GPU is compared to a traditional CPU routine in solving matrices generated by a finite element program.