Towards accelerating smoothed particle hydrodynamics simulations for free-surface flows on multi-GPU clusters

Abstract

Starting from the single graphics processing unit (GPU) version of the Smoothed Particle Hydrodynamics (SPH) code DualSPHysics, a multi-GPU SPH program is developed for free-surface flows. The approach is based on a spatial decomposition technique, whereby different portions (sub-domains) of the physical system under study are assigned to different GPUs. Communication between devices is achieved with the use of Message Passing Interface (MPI) application programming interface (API) routines. The use of the sorting algorithm radix sort for inter-GPU particle migration and sub-domain "halo" building (which enables interaction between SPH particles of different subdomains) is described in detail. With the resulting scheme it is possible, on the one hand, to carry out simulations that could also be performed on a single GPU, but they can now be performed even faster than on one of these devices alone. On the other hand, accelerated simulations can be performed with up to 32 million particles on the current architecture, which is beyond the limitations of a single GPU due to memory constraints. A study of weak and strong scaling behaviour, speedups and efficiency of the resulting program is presented including an investigation to elucidate the computational bottlenecks. Last, possibilities for reduction of the effects of overhead on computational efficiency in future versions of our scheme are discussed.

Composite modelling of subaerial landslide–tsunamis in different water body geometries and novel insight into slide and wave kinematics

“…The computationally most expensive simulation (3D, h = 0.480 m, dp = 10 mm) took 60 h per second of real time and included 23 million particles. Problems involving more than 1000 million particles for free-surface flows have been already investigated with DualSPHysics using the multi-GPU version Valdez-Balderas et al, 2013). Nevertheless, Section 4.2 including a convergence study and Section 4.3 show that dp = 10 mm provides a good compromise between simulation time and accuracy for the present purpose based on the CPU version.…”

Multi-phase SPH modelling of violent hydrodynamics on GPUs

“…Using CUDA as the programming framework for SPH leads to possible confusion with the word 'kernel'. An SPH kernel is the weighting function used in the SPH interpolation process in Equations (5)(6)(7)(8)(9)(10). A CUDA kernel, however, is defined as a C function that is set up and executed N times in parallel by N different CUDA threads.…”

“…Hérault et al [6] and Crespo et al [3] presented comparisons of GPU and CPU codes with significantly improved speedups for the GPU codes. Valdez-Balderas et al [7] and Dominguez et al [8] proposed the use of a multiple GPU system which will allow the simulation of hundreds of millions of particles, up to a billion particles. However, to the best of the authors' knowledge, there has been no study implementing a multi-phase SPH model on GPUs.…”

Second Order Godunov SPH for High Velocity Impact Dynamics

“…An algorithm was proposed to incorporate, into the TS scheme, a constitutive model based on continuum damage mechanics such that the scheme could be used to simulate brittle materials. The TS scheme is more computationally expensive than both the AV and PM schemes (the 2 nd O scheme is approximately 2.5× more expensive than the AV scheme), however advances in massively parallel GPU computing techniques may reduce this burden to the point where the TS scheme is practical [20,36,31,6].…”