Deferred Shading

Deferred shading is a rendering technique that decouples the computation of surface geometry from the computation of lighting, storing intermediate material properties in a geometry buffer (G-buffer) before applying light calculations in a separate pass. Each pixel in the G-buffer stores data such as surface normal, diffuse color, specular intensity, and depth — the raw material facts of the scene, unilluminated.

This decoupling enables scenes with thousands of dynamic lights to be rendered efficiently, because each light only affects pixels it actually illuminates, rather than every object in the scene. The cost is increased memory bandwidth — multiple render targets per pixel — and the loss of native anti-aliasing on transparent or sub-pixel geometry, which the G-buffer cannot represent correctly.

Deferred shading exemplifies a systems principle: when a computation is too expensive to perform in the obvious order, restructure the data flow so that the expensive operation is applied only where necessary. The technique is the dominant lighting architecture in modern game engines and has been extended with variants like tiled deferred shading, which further reduces bandwidth by dividing the screen into tiles and culling lights per tile.