Optimisasi Cloth Simulation Menggunakan Parallel Computing Berbasis GPU pada Platform Nvidia CUDA

Abstract

The advancement in the field of science and technology brings forth new problems and challenges. One such problem is the need to simulate properties of objects or situations due to the real-world conditions being either costly or impossible to achieve. Interactive cloth simulation is one of the fields which requires a simulation to be run in real-time. A realistic simulation requires a heavy computational workload, and as such will hinder the purpose of running the simulation in real-time. Optimization using pure algorithm and math are likely to find balance between accuracy and speed. This research introduces an optimization using parallel computing on GPU to speed up the simulation without reducing the accuracy. The simulation is based on continuum mechanics due to its ability to display realistic physical properties of cloth. The model dynamics is discretized using FEM, then a corotational formulation is used to compute the elasticity property of cloth. This research uses CUDA platform developed by Nvidia to deliver the parallelization. The testing results shows that there are little but acceptable accuracy differences on the parallel version of the simulation. The speedup happens when the structure of a function is ideal for GPU parallelism. A full GPU parallelism causes massive overheads in functions that are not or at least directly parallel compatible. Choosing functions that are compatible or modifying ones to fit the ideal parallel condition will make significant speedup.