G-Buffer Deferred Rendering

What is G-Buffer

G-Buffer (Geometry Buffer) is the core data structure in the deferred rendering pipeline. In deferred rendering, the geometry processing stage does not directly compute lighting; instead, it outputs geometric properties (such as position, normal, albedo, material parameters, etc.) for each pixel into a set of render targets, which is the G-Buffer. Subsequently, the lighting stage uses the data in the G-Buffer to shade each pixel.

G-Buffer is typically composed of multiple render targets (MRT), each storing different attributes.

Typical Composition of G-Buffer

Common G-Buffer layouts include:

• **Albedo**: The base color of objects, usually stored as RGBA8.
• **Normal**: Surface normal in world or view space, often RGBA16F or RGBA8 (compressed format).
• **Position**: World space position of the pixel; can be reconstructed from depth buffer and inverse projection matrix, so sometimes not stored directly to save bandwidth.
• **Material properties**: Such as roughness, metallic, specular reflectance, often packed into one RGBA8 texture.
• **Depth**: Stored in the depth buffer, used for position reconstruction and depth testing.
• **Others**: Such as emissive, AO, custom data, etc.

Different engines have different G-Buffer layouts; for example, Unreal Engine's G-Buffer may include more attributes to support complex materials.

Advantages of Deferred Rendering

• **Decoupled lighting**: Geometry and lighting are separated, allowing efficient handling of many dynamic lights because lighting calculations are per-pixel and independent of scene complexity.
• **Easier post-processing**: Information in the G-Buffer can be used for screen-space effects like SSAO, SSR.
• **Simpler shaders**: Geometry stage uses simple shaders, lighting stage uses a unified lighting model.

Challenges of Deferred Rendering

• **Bandwidth consumption**: G-Buffer requires reading and writing large amounts of data, which can become a bottleneck especially on mobile platforms with limited bandwidth.
• **Memory usage**: Multiple render targets consume video memory.
• **Transparent object handling**: Deferred rendering struggles with transparent objects, often requiring forward rendering assistance.
• **Anti-aliasing**: Traditional MSAA is incompatible with deferred rendering; alternatives like TAA, FXAA are needed.

G-Buffer Optimization

• **Compressed formats**: Use appropriate texture formats to reduce bandwidth, such as storing normals in RGBA8 (via encoding), using R16F for depth, etc.
• **Reducing render target count**: Store only necessary attributes; reconstruct others via computation.
• **Tile-based rendering**: Use tile-based architectures to optimize lighting.
• **Bandwidth optimization**: On mobile, consider forward+ rendering or tile-based deferred.

Summary

G-Buffer is key to deferred rendering. Understanding its design principles and trade-offs helps in developing high-performance real-time rendering applications.