When comparing forward rendering and deferred rendering in 3D game engines, the core difference is the timing of lighting calculations: forward rendering applies lighting during object rendering, while deferred rendering delays lighting until after gathering geometry data.
Forward rendering processes each object and its materials, calculating lighting effects (e.g., diffuse, specular) immediately as the object is drawn. This simplicity makes it easy to implement but becomes inefficient with many dynamic lights, as each light requires reprocessing objects.
Deferred rendering first renders all geometry to a "G-buffer" (storing data like position, normal, albedo), then applies lighting to the entire scene using this precomputed buffer. This approach excels with numerous lights, as lighting is calculated once per pixel rather than per object.
For games with few dynamic lights, forward rendering is practical for its simplicity; for scenes with many lights (e.g., open-world or dynamic environments), deferred rendering better maintains performance.