GPU-Accelerated Particle Physics Simulation

A personal project to practice GPU optimizations using WebGPU and Deno. This project implements a real-time particle physics simulation that can handle thousands of particles with efficient collision detection.

Features

• WebGPU-accelerated particle simulation with CPU fallback
• Grid-based spatial partitioning for efficient collision detection
• Real-time visualization through WebSocket streaming
• Hybrid collision resolution system
• Configurable simulation parameters
• Load testing capabilities
• Unit tests for core components

Tech Stack

• Deno - Modern runtime for JavaScript and TypeScript
• WebGPU - Next-generation graphics and compute API
• WebSocket - Real-time communication between server and client
• TypeScript - Type-safe programming
• WGSL (WebGPU Shading Language) - GPU compute shaders

Project Structure

/
├── back/               # Backend simulation code
│   ├── collision.ts    # Collision resolution logic
│   ├── gpu_simulation.ts # WebGPU simulation implementation
│   ├── grid.ts        # Spatial partitioning
│   ├── particle.ts    # Particle class definition
│   └── shader.wgsl    # GPU compute shader code
├── front/
│   └── index.html     # Frontend visualization
├── test/              # Test suite
└── server.ts          # WebSocket & HTTP server

Getting Started

Prerequisites

• Deno installed on your system
• A browser with WebGPU support (Chrome Canary with appropriate flags)

Running the Project

1. Clone the repository
2. Start the server:

   deno task dev

3. Open http://localhost:8000 in your browser

Running Tests

deno test

Performance

The simulation includes performance testing for different particle counts:

• 1000 particles: Target 1000 FPS
• 5000 particles: Target 150 FPS
• 15000 particles: Target 100 FPS

Implementation Details

• Uses a grid-based spatial partitioning system for collision detection
• Implements both GPU and CPU simulation paths
• WebGPU compute shaders handle particle updates and collisions
• Efficient binary state transfer via WebSocket
• Configurable parameters for mass, radius, and collision elasticity

License

This is a personal project for learning purposes. Feel free to use it as reference.