Home

What is RayON?¶

RayON is an educational and experimental path tracer built in C++ with optional CUDA acceleration. It started as a re-implementation of the classic Ray Tracing in One Weekend series and evolved into a fully interactive raytracer running at > 100 FPS @ 720p on an NVIDIA DGX Spark.

Four rendering back-ends are available at runtime — no recompilation needed:

CPU Single-thread

Reference implementation on the CPU, one pixel at a time. It was the starting point but is in reality useless

CPU Multi-thread

Splits the image into tiles and dispatches them across all available cores using std::async. Typical speedup: 8–16×, depending on your CPU.

CUDA GPU

One-shot CUDA kernel with 32×4 thread blocks, warp-friendly memory layout, and persistent curand states. ~100–500× faster than single-thread CPU.

CUDA Interactive

SDL2 window with progressive accumulation. Orbit, pan, zoom with the mouse. Dear ImGui sliders for live DOF, samples, light intensity, and roughness.

It also features BVH Acceleration, CPU-built, with GPU-traversed Bounding Volume Hierarchy with Surface Area Heuristic (SAH) splitting. This provides 5–50× speedup on scenes with 100+ objects.

Quick start¶

# Build (requires CMake ≥ 3.20, a C++17 compiler, and optionally CUDA + SDL2)
mkdir -p build && cd build
cmake .. --fresh
make -j$(nproc)

# Run
./rayon
# > Choose renderer: 0=CPU  1=CPU-parallel  2=CUDA  3=CUDA interactive

Load one of the bundled example scenes:

./rayon --scene ../resources/scenes/09_color_bleed_box.yaml -s 512 -r 1080

See Getting Started for the full setup guide, or YAML Scene Format to author your own scenes.

Explore the docs¶

Section	What you'll find
How It Works	The math: ray equations, material models, BVH, sampling theory
Architecture	Code organization, CUDA renderer internals, progressive pipeline
Features	Interactive controls, YAML scene format, SDF shapes, OBJ loading
Gallery	Curated renders from all available scenes
Performance	Benchmark results, speedup tables, tuning tips