Fast Fluid Simulations with Sparse Volumes on the GPU

Abstract

We introduce efficient, large scale fluid simulation on GPU hardware using the fluid‐implicit particle (FLIP) method over a sparse hierarchy of grids represented in NVIDIA® GVDB Voxels. Our approach handles tens of millions of particles within a virtually unbounded simulation domain. We describe novel techniques for parallel sparse grid hierarchy construction and fast incremental updates on the GPU for moving particles. In addition, our FLIP technique introduces sparse, work efficient parallel data gathering from particle to voxel, and a matrix‐free GPU‐based conjugate gradient solver optimized for sparse grids. Our results show that our method can achieve up to an order of magnitude faster simulations on the GPU as compared to FLIP simulations running on the CPU.

Fast Nonlinear Least Squares Optimization of Large‐Scale Semi‐Sparse Problems

“…The problem may also be sparse and defined on a regular domain, like an image, a grid, or a mesh, i.e., they have a local nature. In these cases, it is possible to write fast, dedicated GPU solvers, which quickly solve the particular problem at hand but may be difficult to generalize, e.g., in fluid simulation [LMAS16, WTYH18, CZY17], and shape reconstruction [DNZ∗17]. If the topology of the system does not change, then the solver can even be automatically created by using a domain specific language, such as OptLang [DMZ∗17].…”

Multi‐Level Memory Structures for Simulating and Rendering Smoothed Particle Hydrodynamics

“…Furthermore, various CPU-based approaches exist, for example, OpenVDB [MLJ*13], but they often require significant changes to be realized on GPUbased systems. OpenVDB was realized for GPUs as GVDB, where recently, Wu et al [WTYH18] introduced a GVDB-based data structure for FLIP-based simulations that significantly improves performance, but which is not directly applicable to SPH, as FLIP imposes significantly different requirements on the data structure, which is an integral part of the simulation itself. Overall, none of the existing data structures can be applied directly both for rendering and simulation without significant overhead.…”

An Efficient Geometric Multigrid Solver for Viscous Liquids

“…This approach was demonstrated both by Liu et al [2016], who advocated a hybrid GPU/CPU approach, and Chu et al [2017], who applied a recursive decomposition and distinct solver types on the resulting sub-domains. Other common but largely orthogonal techniques for accelerating fluid solvers include h-or p-adaptivity (e.g., [Edwards and Bridson 2014;Losasso et al 2004]) and GPU implementation [Bolz et al 2003;Wu et al 2018]. Most recently, Goldade et al [2019] proposed a symmetric, variational octree extension of the method of Batty and Bridson [2008].…”