FLUID SIMULATION BASED ON MPM ON GPU USING CUDA

• Main Author: FAIZLANI KUSNADI - NIM 13512054 , LUQMAN
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/28662
• Institution: Institut Teknologi Bandung

To produce realistic simulations, fluid simulation requires large amounts of particles and high framerate. One way to produce a high framerate with a large number of particles is parallelization. This Final Project discusses the implementation of fluid simulation on Graphics Processing Unit (GPU) with Compute Unified Device Architecture (CUDA) to improve fluid simulation performance. <br /> <br /> <br /> <br /> <br /> The simulation algorithm used is Material Point Method (MPM) to facilitate the arrangement of simulated fluid material. Implementation on the GPU is done by parallelizing the simulation code and will run on CUDA. Parallelization is done by dividing the simulation into six kernels that will run serially, and each kernel runs in parallel. <br /> <br /> <br /> <br /> <br /> MPM fluid simulation implementation is done using the Cinder library with the C++ language. The results of the implementation of MPM fluid simulation on the GPU using CUDA shows better performance than the serial version. With a performance increase of around 1.5-2.4 times on Nvidia Geforce GTX 950m and 4.1-6.6 times on Nvidia Titan Xp.