Mesh Shaders and Primitive Shaders: Replacing Fixed-Function Vertex Pipelines: Fundamental Mechanics and Deep Mathematical Modeling

[Hunter Cook]

I’ve been digging into how mesh shaders actually replace the fixed-function vertex pipeline, and I want to understand the architecture, not just the API surface. When geometry amplification runs in the mesh shader, where exactly does the workload sit relative to the warp scheduler—does the amplification factor get dispatched as separate primitives before rasterization, or is it handled as a compute-style wavefront that later feeds the primitive assembler? And in practice, how do you handle dynamic LOD culling without stalling the pipeline? I’m curious whether people are doing frustum/occlusion culling inside the mesh shader itself or keeping it in a pre-pass, and what the overhead looks like when you’re culling per-object versus per-submesh.

On the math side, I’m trying to reconcile how micro-polygon rasterization interacts with the new pipeline stages. When you generate dense geometry, how do you model the fill-rate cost versus the geometry-shader savings, and is there a clean way to express the throughput ceiling before you start losing the vertex cache advantage? I’ve seen the theoretical proofs for primitive throughput scaling, but the real-world numbers seem to diverge a lot depending on shader complexity. Anyone here benchmarked mesh shader pipelines against traditional geometry shaders on competitive titles, and what were the actual frame-time deltas at high refresh rates?

"Half-Life, Deus Ex, and System Shock 2: Peak PC gaming era."

[Author]

Posts