Rendering — System Overview

An index of the rendering subsystems and the design thinking behind them.

Goals

Zero CPU overhead. All culling on the GPU. No per-instance CPU work.
Scales to lower-end hardware. Minimum spec is the Steam Deck. Mesh shaders need a GTX 1650; a vertex shader fallback is planned for later.
Realtime all the time. No light baking, no overnight precomputation.
Stable, clean image. Motion clarity and stability over extreme fidelity or photorealism. TAA is strictly optional — we evaluate image quality without it.
Native resolution. Upscaling is strictly optional. Everything is designed to run at native.
Technical-artist friendly. Write custom shaders, modify the lighting code, change parameters at runtime, access command buffers directly.
Low latency. Double buffering by default; triple buffering in build config.

We make a few deliberate bets to hit those goals:

GPU-driven. Instance culling, cluster culling, and indirect draw generation all run in compute.
Bindless. Every shader sees every resource, every resource is a 16-bit handle.
Mesh shaders. No vertex buffers, no input layouts. Geometry arrives as compressed clusters, decompressed one thread per vertex.
Strict shader permutation control. No material graphs, no hidden shader variants.