CUDA Ray TracerHigh-Performance GPU Ray Tracer

A technical whitepaper on real-time ray tracing with CUDA, featuring BVH acceleration, Monte Carlo path tracing, and warp divergence optimization

Architecture Overview

Get Started

GitHub

Rendering Results

Blinn-Phong Shading800×600 · <1s render time

Path Tracing (Cornell Box)800×600 · 128 samples

Random Spheres800×600 · BVH accelerated

Technical Highlights

🌳 BVH Acceleration

O(log N) intersection queries via hierarchical bounding boxes. Stack-based GPU traversal with 64-element depth limit.

🎲 Monte Carlo Path Tracing

Physically-based global illumination with cosine-weighted hemisphere sampling and Russian roulette termination.

⚡ Ray Sorting Optimization

Thrust-based sort-by-key grouping rays by hit object ID. Reduces warp divergence by 20-40%.

📦 GPU Memory Model

Optimized memory layout with coalesced access, shared memory BVH stack, and SoA data structures.

🎨 Material System

Extensible material presets: matte, plastic, metal, mirror. Custom BRDF support via virtual interface.

🖼️ Tone Mapping

Reinhard and ACES filmic tone mapping operators for HDR to LDR conversion with gamma correction.

Quick Start

Build and run your first render in 5 minutes

 # Clone the repository
 git clone https://github.com/AICL-Lab/ray-tracer.git
 cd ray-tracer

 # Build (requires CUDA 12+)
 cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
 cmake --build build -j$(nproc)

 # Render (Phong shading)
 ./build/bin/ray_tracer -o output.ppm

 # Render (Path tracing)
 ./build/bin/ray_tracer --mode path-trace --samples 128 -o output.ppm