DualSPHysics: When Your Fluid Doesn’t Want to Stay in a Mesh
Some flows don’t cooperate with a mesh. A wave breaking over a seawall, a dam failing catastrophically, debris tumbling through floodwater, a tank sloshing under motion — these are all problems where the free surface fragments, merges, and moves through large deformations that make traditional mesh-based CFD expensive or brittle to set up. DualSPHysics takes a different approach entirely: it doesn’t use a mesh at all.
What DualSPHysics is
DualSPHysics is an open-source, GPU-accelerated solver based on Smoothed Particle Hydrodynamics (SPH) — a meshless, Lagrangian method that represents the fluid as a set of moving particles rather than field values on a fixed grid. Each particle carries its own mass, velocity, and pressure, and interacts with nearby particles through a smoothing kernel; the flow’s behavior emerges from tracking how these particles move and interact over time, rather than from solving equations on cells that the fluid has to pass through.
The project originated at the University of Vigo and the University of Manchester, building on the earlier SPHysics FORTRAN code, and has been developed continuously since its first release in 2011. It’s written in C++ and CUDA, was conceived specifically for coastal engineering problems — wave impact on structures being a founding use case — and has since grown into a general-purpose solver used across coastal, hydraulic, and industrial free-surface flow problems, with adoption spanning academic research groups and industrial users alike.
Why meshless matters here
The appeal of SPH isn’t that it’s better than mesh-based CFD everywhere — it isn’t. It’s that certain classes of problems are naturally suited to a particle representation:
- Violent free-surface flows. Dam breaks, wave breaking and overtopping, sloshing, and green-water loading on ships all involve the free surface fragmenting into droplets, merging back together, and deforming past anything a body-fitted mesh could track without constant remeshing. A particle method simply doesn’t need the surface to stay connected.
- Fluid-driven and multi-body interaction. DualSPHysics couples its SPH fluid solver with a Discrete Element Method (DEM) for rigid body interactions, so debris, floating structures, or colliding bodies driven by the fluid are handled within the same framework rather than bolted on afterward.
- Coupling with wave propagation models. For large coastal domains, DualSPHysics can be coupled with models like SWASH or OceanWave3D, using the efficient wave-propagation solver for the open ocean and switching to SPH only in the nearshore zone where the violent, detailed physics actually happens.
- Extending beyond fluids. Recent extensions in the DualSPHysics ecosystem — including solid-mechanics formulations for hyperelasticity, plasticity, and fracture, and large-deformation geomechanics solvers for landslides and geomaterial flows — build on the same particle framework, reflecting how far the method has been pushed beyond its original coastal-engineering scope.
Built for the GPU from the ground up
This is the part that matters most for how you’d actually run it. SPH is, by its structure, an unusually GPU-friendly method: the particle-interaction computation that dominates runtime is inherently parallel, since each particle’s update depends only on a local neighborhood rather than a global linear system. DualSPHysics was engineered around this from early on, with a CUDA implementation alongside its OpenMP-based CPU path, and published performance studies have consistently shown large speedups moving from CPU to GPU execution on both dam-break benchmarks and full-scale coastal cases.
That GPU-native design is also why DualSPHysics scales to problems that would be impractical otherwise. Real coastal and industrial cases studied in the literature — tsunami run-up modeling of full urban districts, large-scale geomechanics problems with on the order of 10⁸ particles on a single GPU — are only tractable because the solver was built to use accelerator hardware as its primary execution path, not as an afterthought bolted onto a CPU codebase. Multi-GPU support is also on the near-term roadmap, expected as a major feature of the upcoming v6.0 release.
Where cloudHPC fits in
A GPU-first solver is only as useful as the GPU you can point it at, and that’s precisely the constraint DualSPHysics runs into on a typical workstation: a single consumer or workstation GPU caps the particle count and domain size you can realistically simulate, and buying dedicated high-end GPU hardware for work that spikes around specific project phases is hard to justify for most engineering teams.
On cloudHPC, DualSPHysics runs on GPU-equipped instances provisioned specifically for this kind of workload — sized to the case rather than to whatever happens to be sitting in a workstation. A quick dam-break validation run doesn’t need the same resources as a full-scale tsunami run-up study or a wave-impact assessment with millions of particles and DEM-coupled floating debris; on-demand provisioning means the compute matches the problem, not the other way around. And because SPH studies are frequently run as scenario sweeps — different wave conditions, different structure geometries, different storm return periods — the ability to spin up multiple GPU instances in parallel turns a sequential batch of long runs into a set of simultaneous ones.
If your work involves wave-structure interaction, dam-break and flood modeling, coastal defense design, or any free-surface problem where a mesh becomes more trouble than it’s worth, that’s the gap cloudHPC is built to close.
Want to run DualSPHysics without provisioning your own GPU hardware? cloudhpc.cloud provides on-demand GPU instances pre-configured for DualSPHysics alongside CFD and FEA simulation workloads.
CloudHPC is a HPC provider to run engineering simulations on the cloud. CloudHPC provides from 1 to 224 vCPUs for each process in several configuration of HPC infrastructure - both multi-thread and multi-core. Current software ranges includes several CAE, CFD, FEA, FEM software among which OpenFOAM, FDS, Blender and several others.
New users benefit of a FREE trial of 300 vCPU/Hours to be used on the platform in order to test the platform, all each features and verify if it is suitable for their needs
