Variable Rate Shading (VRS) Tier 2 in High-Framerate Competitive Titles: Advanced Implementation and Real-World Architecture

[Brandon Howard]

Been pushing VRS Tier 2 (per-shading-group 8×8 granularity) in CS2 and Valorant to shave latency while holding 360fps, but I’m hitting a wall on how the driver actually maps shading rates to the rasterizer pipeline under sustained motion. When you’re sweeping the mouse fast, does the tier-2 mask get recomputed per-frame, or is there a fixed region budget that gets reloaded? And how much of the peripheral artifacting I’m seeing around muzzle flashes is actual coarse-shading blur versus the compositor’s blending of mismatched shading regions?

For anyone running engine-level configs: in Unreal’s foveated rendering pass, are you driving the shading rate via the render target’s MIP chain or through the pipeline state object’s VRS register? How does that interact with the fixed-function early-Z stage—does shading peripheral tiles at a reduced rate cause tile-boundary popping when the view matrix shifts between sub-frames? I’ve got benchmark data from RTX 40-series and RDNA3 cards showing 8-12% frame time spikes when tier-2 regions cross tile boundaries, but I want to know if you’re seeing the same or if your driver version handles the mask differently.

"~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~"

"Veni, vidi, vici." • Esports & PC gaming enthusiast"

[Author]

Posts