Parallel Graphics Software Design

Abstract

Many grand challenge applications such as weather prediction are known for their extensive computations and large datasets, and are usually run on massively parallel processor systems (MPPs). To visualize data from these applications, potentially massive amounts of data must be transferred from the MPPs to special graphics workstations, often across a network. Under this condition, it may be preferable to visualize the data in place. In this paper, we first reveal the usefulness of directly perform rendering on parallel machines, and then discuss design issues related to this field. Finally, we present a sort-last class polygon rendering method which performs better than our previous work by 13-30% times faster. To the best of our knowledge, the reported rendering rate obtained from a 512 node Intel Delta, is much higher than any other parallel implementation on the same kind or similar parallel architectures.