Overview
This project is a real-time 3D fluid simulation using the Smoothed Particle Hydrodynamics (SPH) method. Implemented from scratch in C++, the simulation models fluid behavior by representing water as a collection of particles that interact through pressure, viscosity, and external forces. Particles are accelerated using force computations derived from SPH equations and integrated over time using a stable numerical solver.
To optimize neighbor searches, the simulation uses a Spatial Hash Grid, enabling efficient retrieval of nearby particles per frame. The simulation runs at 60 FPS with 1000 particles on a single-threaded CPU. The simulation is integrated into “Bark” my own physics engine.
WORK IN PROGRESS: I’m currently working on a compute shader version of the simulation to significantly improve performance and support a higher number of particles.
How it looks like
Here is a gif of the CPU simulation:
What did I learn
-
Smoothed Particle Hydrodynamics (SPH): Gained a deep understanding of SPH principles, including density estimation, pressure and viscosity forces.
-
Spatial Hash Grid: Implemented an efficient spatial partitioning system to accelerate neighbor searches in large particle sets.
-
Numerical Stability: Learned to prevent instability in particle simulations by carefully tuning parameters and using stable integration methods.
-
Performance Profiling: Analyzed and optimized CPU performance to maintain real-time frame rates.
-
Parallel Computation: Began transitioning physics computations to GPU via compute shaders to support larger-scale simulations.
-
Engine Integration: Embedded the simulation within my own physics engine “Bark”, improving modularity and system-level understanding of engine architecture.