high performance computing on graphics processing units: hgpu.org

PURPOSE: The purpose of this paper is to determine the possibility of implementing parallel processing feature of graphic processor unit (GPU) in garment drape simulation. DESIGN/METHODOLOGY/APPROACH: Velocity-Verlet method based on explicit integration is used to drape triangular table cloth meshes. Both drape simulation and collision detection engines are converted to GPU version. Simulation speeds of simple linear algebra and actual free fall table cloth simulation are compared with those of the central processing unit (CPU) version. FINDINGS: There is apparent calculation speed increase when the parallel computation of GPU is implemented. But the current GPUs have several limits for general purpose computation, so the original CPU version algorithm should be split and modified to be used in GPU. ORIGINALITY/VALUE: This paper implemented GPU parallel processing technique in both drape simulation and collision detection. Voxel-based method is used to find possible collision pairs. Triangular meshes, which are more difficult to implement than quadrilateral ones in GPU programming, are successfully implemented.