Fast parallel index-based GPU particle collision detection for science and engineering.

Abstract A fast particle collision detection method by rounding and indexing particle and cell locations. The two part process is performed in compute and render pipelines exclusively on the GPU. The method is GPU-Driven in that detection and rendering are simultaneous and all logic resides on the GPU. The application can process 1 million particles in less than 1/30th of a second, (30 frames per second) on an inexpensive laptop workstation. This speed provides a basis for scientists and engineers accustomed to computational fluid dynamics, to consider particle systems for thremo-fluid flow including rarefied gases. It presents an opportunity for students and educators to explore massive particle systems (greater than 1 million particles). Also, in order to encourage comparative analysis, we have developed a light-weight benchmarking method where test data is generated by software.