Anyone who plays video games or who goes to movies that use computer-generated imaging knows that virtual environments are becoming more and more realistic. Also, computers available to consumers continue to get more powerful, with multiple cores. The algorithms that use these computers to create virtual environments need to be structured in such a way that they can create realistic, interactive realms in games and other virtual-reality applications as efficiently as possible.
One way to improve efficiency is to create a parallel program, which is broken into several pieces that are run concurrently, instead of each piece running sequentially. Software developers of video games are working to develop programs that can be parallelized in order to make use of multiple cores or computer clusters. MSI Principal Investigator Stephen Guy, an assistant professor in the Department of Computer Science and Engineering (College of Science and Engineering), and members of his group are working on these kinds of programs. At the 6th International Conference of Motion in Games in Dublin, Ireland, in November 2013, Professor Guy, along with student John Koenig and post-doctoral researcher Dr. Ioannis Karamouzas, presented a paper discussing a new object-centric algorithm for parallel rigid-body simulation. The object-centric method means that each simulated body is modeled independently and is therefore self-contained. Objects that are more isolated from other objects can be modeled with larger time-steps, which improves efficiency and saves computation time for objects that are interacting more closely.
This object-centric method results in interactive, real-time simulations that can scale across many CPU cores. This paper included scenarios that consisted of hundreds of interacting objects. The paper can be found at the Association for Computer Machinery’s Digital Library (Koenig, John, Ioannis Karamouzas, Stephen J. Guy. Object-centric parallel rigid body simulation with timewarp. 2013. Proceedings of MIG ’13, Motion of Games, 203-212. DOI: 10.1145/2522628.2522652).
Image description: A dynamic scene with two hundred spheres falling onto five static cylinders. This simulation approach is object-centric, with each modeled as a soft-thread and simulated independently. This results in scalable performance, achieving a 5-6x simulation speedup on eight cores and a 9-10x speedup on 16 cores. Image and description J. Koenig et al., Proceedings of MIG ’13, Motion of Games, 203-212 (2013). ©2013 Association for Computer Machinery
posted on September 3, 2014