Skip to content

Architecture

This section explains how RayON is structured internally — how the scene is built, how the CUDA kernel launches work, and how the progressive renderer accumulates samples.


Scene System

The "build once, render anywhere" design: one neutral SceneDescription converted into CPU class hierarchy or GPU flat structs.

CUDA Renderer

32×4 thread blocks, thread→pixel mapping, curand states, float accum buffer, and the C++/CUDA boundary.

Progressive Rendering

The accumulation loop, tiled rendering, sample stages, adaptive depth, and live parameter injection.