Overlapping multi-processing and graphics hardware acceleration: performance evaluation

Recently, multi-processing has been shown to deliver good performance in rendering. However, in some applications, processors spend too much time executing tasks that could be more efficiently done through intensive use of new graphics hardware. We present a novel solution combining multi-processing and advanced graphics hardware, where graphics pipelines are used both for classical visualization tasks and to advantageously perform geometric calculations while remaining computations are handled by multi-processors. The experiment is based on an implementation of a new parallel wavelet radiosity algorithm. The application is executed on the SGI Origin2000 connected to the SGI InfiniteReality2 rendering pipeline. A performance evaluation is presented. Keeping in mind that the approach can benefit all available workstations and super-computers, from small scale (2 processors and 1 graphics pipeline) to large scale (p processors and n graphics pipelines), we highlight some important bottlenecks that impede performance. However, our results show that this approach could be a promising avenue for scientific and engineering simulation and visualization applications that need intensive geometric calculations.