A massively parallel and scalable multi-GPU material point method

Abstract

Harnessing the power of modern multi-GPU architectures, we present a massively parallel simulation system based on the Material Point Method (MPM) for simulating physical behaviors of materials undergoing complex topological changes, self-collision, and large deformations. Our system makes three critical contributions. First, we introduce a new particle data structure that promotes coalesced memory access patterns on the GPU and eliminates the need for complex atomic operations on the memory hierarchy when writing particle data to the grid. Second, we propose a kernel fusion approach using a new Grid-to-Particles-to-Grid ( G2P2G ) scheme, which efficiently reduces GPU kernel launches, improves latency, and significantly reduces the amount of global memory needed to store particle data. Finally, we introduce optimized algorithmic designs that allow for efficient sparse grids in a shared memory context, enabling us to best utilize modern multi-GPU computational platforms for hybrid Lagrangian-Eulerian computational patterns. We demonstrate the effectiveness of our method with extensive benchmarks, evaluations, and dynamic simulations with elastoplasticity, granular media, and fluid dynamics. In comparisons against an open-source and heavily optimized CPU-based MPM codebase [Fang et al. 2019] on an elastic sphere colliding scene with particle counts ranging from 5 to 40 million, our GPU MPM achieves over 100x per-time-step speedup on a workstation with an Intel 8086K CPU and a single Quadro P6000 GPU, exposing exciting possibilities for future MPM simulations in computer graphics and computational science. Moreover, compared to the state-of-the-art GPU MPM method [Hu et al. 2019a], we not only achieve 2x acceleration on a single GPU but our kernel fusion strategy and Array-of-Structs-of-Array ( AoSoA ) data structure design also generalizes to multi-GPU systems. Our multi-GPU MPM exhibits near-perfect weak and strong scaling with 4 GPUs, enabling performant and large-scale simulations on a 1024 3 grid with close to 100 million particles with less than 4 minutes per frame on a single 4-GPU workstation and 134 million particles with less than 1 minute per frame on an 8-GPU workstation.

A Novel Plastic Phase‐Field Method for Ductile Fracture with GPU Optimization

“…On the other hand, in order to verify the effectiveness of our GPU optimization strategy, we have already presented two benchmark tests for our strategy in §5. Since our approach accelerates the MPM preprocessing and phase‐field evolution, it can be combined with recent acceleration methods for MPM workflow in [WQS*20, GWW*18] to further improve the efficiency. Besides, we benchmark the overall PPF‐MPM and traditional MPM in Fig.…”

“…In this case, we tend to choose the non‐associative flow rule to ensure the interaction performance and algorithm robustness, and satisfy the requirements of graphical applications in our future work. Furthermore, AoSoA and G2P2G [WQS*20] can be applied in our GPU framework to further decrease the time consumption of MPM workflow, so as to improve the speedup rate of the entire MPM algorithm.…”

“…In GPU MPM, Gao et al [GWW*18] optimized the grid‐particle transfer and particle reordering based on SPGrid, and also provided specific strategies for the numerical calculation, such as the GPU singular value decomposition (SVD) solver. In the latest GPU‐based MPM [WQS*20], Wang et al proposed G2P2G and AoSoA with multiple GPUs, with which the modified MPM workflow has been greatly accelerated. In this case, the time cost of processes like reinitialization cannot be ignored compared with the total time of an MPM step.…”

QuanTaichi

“…Work in graphics explores high-resolution simulation on multicore CPUs [Aanjaneya et al 2017;Liu et al 2018Liu et al , 2016McAdams et al 2010;Setaluri et al 2014] and massively parallel GPUs [Gao et al 2018;Wang et al 2020;Wu et al 2015. Corresponding sparse data structures are proposed to support the underlying structured grid, often with a certain degree of bit-compression [Hoetzlein 2016;Houston et al 2006;Museth 2013;Setaluri et al 2014].…”

“…Our system practically pushes the limit of simulation resolutions by alleviating the memory space constraints. For example, with the help of quantization, our single GPU MLS-MPM ] simulation of 235 million particles, has a higher resolution than the existing highest resolution MPM simulation on 8 GPUs (134 million particles, see [Wang et al 2020]).…”