CUDA-Accelerated Ray Tracer with Audio-Reactive Boids

A real-time ray tracer implemented in Python and CUDA that combines physically-based rendering with an audio-reactive birds flocking simulation - using the Boids algorithm.

Note: This is my tiny tribute to one of my favourite songs - Flume (fav artist) X Greenpeace (environmental activism network)

Features

A Basic Ray Tracing Engine

• CUDA kernel for ray-sphere intersection and shading calculations
• Lighting with ambient, diffuse, and specular shading
• Soft shadows and reflective surfaces
• Real-time perf through GPU acceleration
• Shitty Aim Trainer using PyGame for ui and coloring the pixels.

Audio-Reactive Boids Simulation

• Implementation of Craig Reynolds' Boids flocking algorithm
• 256 concurrent bird agents rendered as reflective spheres
• Real-time audio reactivity:
  • Flock behavior modulated by frequency analysis
  • Movement parameters controlled by beat detection
  • Visual properties (color, size, reflectivity) respond to audio features

FPS Controls

• First-person camera controls (WASD + mouse)
• Real-time spectator mode to observe the flock
• Dynamic adjustment of simulation parameters

Performance

• Significant speedup achieved through CUDA acceleration (1000x if my calulations are right)
• Comparative benchmarks between:
  • Raw Python implementation
  • Numba JIT compilation
  • CUDA parallel processing
• Real-time performance monitoring and visualization

Implementation Details

CUDA Optimizations

• Custom CUDA kernels for core rendering ops

Audio Processing

• Real-time FFT analysis for frequency separation
• Onset detection for beat recognition
• Multi-band audio feature extraction

Getting Started

Prerequisites

• Python 3.8+
• NVIDIA GPU with CUDA support
• Poetry for dependency management

Installation

# Clone the repository
git clone https://github.com/floaredor/cuda-raytracer.git
cd cuda-raytracer

# Install dependencies
poetry install

Running the Demo

# Run with default audio file
python experiments/music_boids.py

# Run with custom audio file
python experiments/music_boids.py path/to/your/audio.wav

Controls

• W/A/S/D - Move camera
• Mouse - Look around
• ESC - Exit and generate performance plots

Visualization Examples

Shadows

Boids Algorithm

Random Experimental

Glazed Surfaces

To Do

• Multiple light sources
• Anti-aliasing
• More complex geometries
• UI for parameter adjustment like ImGUI

Project Background

This project was developed as a practical exploration of NVIDIA's CUDA Python course concepts. It shows the application of parallel processing techniques to graphics and nature-based simulation while incorporating real-time audio processing for an interactive experience.

License

MIT

Acknowledgments

• NVIDIA for their CUDA Python course
• Craig Reynolds for the original Boids algorithm