Deferred Rendering

In 'Deferred' rendering, on the other hand, we defer the shading and blending of light information until after a first pass over the screen where positions, normals, and materials for each surface are rendered to a ‘geometry buffer’ (G-buffer) as a series of screen-space textures. We then composite these results together with the lighting pass. This approach has the principle advantage that the render cost of lighting is proportional to the number of pixels that the light illuminates, instead of the number of lights themselves. As a result you are no longer bound by the number of lights you wish to render on screen, and for some games this is a critical advantage.

Deferred Rendering gives highly predictable performance characteristics, but generally requires more powerful hardware. It is also not supported by certain mobile hardware.

For more information on the Deferred, Forward and the other available rendering paths, please see the documentation here.