Mesh Shaders and Primitive Shaders: Replacing Fixed-Function Vertex Pipelines: Competitive Benchmark Analysis and 硬件 Interactions

[Christian Ramirez]

I’ve been running empirical benchmarks across Mesh Shader workloads in a few competitive titles, pushing geometry amplification factors and dynamic LOD culling to see where the fixed-function vertex pipeline actually breaks down. Before I commit to a full comparison across the Turing and RDNA generations, I want to hear from people who’ve actually stress-tested these in real matches. At what geometry amplification multiplier do you start seeing frametime spikes, and does that threshold differ meaningfully between card vendors?

Also curious about your testing methodology—how are you isolating GPU-bound frametime inconsistency from driver-level overhead? When you enable micro-polygon rasterization and aggressive culling, do you notice the LOD transitions popping mid-fight, or does the transition stay smooth enough to not break your aim? And for anyone benchmarking at high refresh rates, how much does hardware scaling actually improve frametime consistency versus raw frame count? I’m looking for concrete numbers and settings, not marketing specs.

[Author]

Posts