GBuffer

A G-buffer, or geometry buffer, is a set of per-pixel data produced during the first pass of a deferred shading pipeline. In this approach, scene geometry is rendered once to populate several render targets with information about each visible fragment. The data stored in a G-buffer typically includes position (in view or world space), normal vector, albedo (diffuse color), and material properties such as specular color, roughness, and metalness, as well as an ambient occlusion term and a depth value. Some implementations also store material IDs or emissive data, and data may be packed to save memory.

During the lighting pass, lighting calculations are performed per pixel by sampling the G-buffer and combining the stored data with scene lights. This decouples geometry processing from shading, allowing many lights to be applied efficiently without re-sending geometry to the GPU.

Implementations usually use multiple render targets (MRTs) to write the various buffers in a single pass. A separate depth buffer is often maintained for depth testing and post-processing. Trade-offs include reduced shading cost for scenes with many lights, at the expense of increased memory bandwidth and complexity. Handling transparency is challenging in deferred rendering, since transparent surfaces typically require forward rendering or special handling. Additionally, post-processing and screen-space effects must account for the G-buffer layout. Some engines pack data to reduce bandwidth, and newer approaches such as Forward+ or tiled shading mix concepts to address certain limitations.