high performance computing on graphics processing units: hgpu.org

Providing realistic, high-resolution and high fidelity representation of motions ia essential in the cloth simulation problem. In order to make high resolution simulations tractable, several algorithms have been developed that manage cloth-object interactions efficiently through specialized data structures such as AABB trees. However, implementation restrictions on single CPU architectures impose certain limits on quality and performance in high-demanding simulations, motivating the study of new implementation techniques. In this paper we address several critical issues in high resolution cloth simulation, enabling us to represent and simulate intricate folds and wrinkles. We employ AABB hierarchies to optimize detection and response in cloth-object collisions. By employing a multi-processor approach on multi-threaded CPU and an emerging multi-core GPU-CUDA architecture, we quantitatively evaluate the workload and computational effort of the cloth simulation application. In addition to this quantitative performance evaluation on multi-processor architectures we illustrate the potential of our approach by presenting a variety of high-quality and high-resolution simulations of cloth behavior under different cloth-object interactions.