Mesh shading restructures the traditional vertex–geometry pipeline by moving amplification, culling, and primitive generation into a GPU-driven task/mesh stage. This seminar paper investigates how meshlet-based rendering and cluster culling can improve real-time performance for highly detailed scenes. We outline the mesh shader execution model, discuss practical work distribution between task and mesh shaders, and compare common visibility tests such as frustum, backface, and depth-based occlusion culling on a meshlet granularity. Building on prior work on conservative bounds for skinned meshes and depth-based visibility, we highlight robustness concerns (e.g., animation-induced bound invalidation) and implementation constraints imposed by current APIs and hardware. Finally, we summarize industry adoption patterns and derive guidelines for integrating mesh shaders into modern engines, with a focus on achieving predictable performance while maintaining correctness.
Defne Göncü (Thu,) studied this question.