Rigged LOD triggers: memory-driven reduction as VRAM ladder steps; perf trigger only on real rigged GPU time

sky_line commented

2026-10-10 01:15:50 +00:00

Owner

The rigged controller (#559 / #561, rigPressure) has two inputs today:

memory: predicted visible rigged vertices over ~85% of the rigged budget, which itself shrinks with the VRAM ladder;
perf: smoothed frame time slower than RenderVKRiggedLodTargetFPS (60) and visible skinned vertices over a per-card budget (0.6M per GB of VRAM budget).

Changes

Memory-driven reduction becomes part of the VRAM ladder: decimated steps, then creator-LOD caps, as ordered rungs in the ladder's own sequence (with its hysteresis and recovery), not a parallel controller. The ladder decides what to give up first; on <= 8 GB cards it should take from textures before shrinking the rigged budget (mid-texture blur looks better than bodies poking through clothes).
Perf trigger only on real rigged GPU cost: engage on the measured GPU time of the rigged passes (pre-skin + rigged raster; the gpuSubMark timers) against a budget, not on whole-frame time. A CPU-bound frame (e.g. a Ryzen 7 2700 at 10-15 FPS in a club) no longer pays a quality cost for nothing.
Log lines name which trigger engaged.

Related: #559, #561, the clothing-shrinkage ticket.

The rigged controller (#559 / #561, `rigPressure`) has two inputs today: - **memory:** predicted visible rigged vertices over ~85% of the rigged budget, which itself shrinks with the VRAM ladder; - **perf:** smoothed frame time slower than RenderVKRiggedLodTargetFPS (60) and visible skinned vertices over a per-card budget (0.6M per GB of VRAM budget). ## Changes - **Memory-driven reduction becomes part of the VRAM ladder**: decimated steps, then creator-LOD caps, as ordered rungs in the ladder's own sequence (with its hysteresis and recovery), not a parallel controller. The ladder decides what to give up first; on <= 8 GB cards it should take from textures before shrinking the rigged budget (mid-texture blur looks better than bodies poking through clothes). - **Perf trigger only on real rigged GPU cost**: engage on the measured GPU time of the rigged passes (pre-skin + rigged raster; the gpuSubMark timers) against a budget, not on whole-frame time. A CPU-bound frame (e.g. a Ryzen 7 2700 at 10-15 FPS in a club) no longer pays a quality cost for nothing. - Log lines name which trigger engaged. Related: #559, #561, the clothing-shrinkage ticket.

sky_line added the

labels

2026-10-10 01:15:50 +00:00

sky_line added this to the Slipstream project

2026-10-10 01:15:50 +00:00

sky_line added

status

in-progress

and removed

triage

labels

2026-10-10 01:24:23 +00:00

sky_line referenced this issue

2026-10-10 01:28:56 +00:00

Mesh upload: better auto-generated LODs (attribute / skin-weight aware simplification, vertex placement, error shown in the UI) #618

sky_line referenced this issue from a commit

2026-10-10 03:31:53 +00:00

#613: rigged LOD triggers -- memory reduction as VRAM-ladder rungs, perf trigger on measured rigged GPU time

sky_line commented

2026-10-10 05:33:00 +00:00

Author

Owner

Built and pushed (d697fff684): memory reduction is now rungs of the VRAM ladder and the perf trigger only fires on measured rigged GPU time. Shipped in 1.2.0. Not yet verified in-world on a heavy venue / small-VRAM card, so it moves to Testing until that is confirmed.

Built and pushed (d697fff684): memory reduction is now rungs of the VRAM ladder and the perf trigger only fires on measured rigged GPU time. Shipped in 1.2.0. Not yet verified in-world on a heavy venue / small-VRAM card, so it moves to Testing until that is confirmed.

sky_line added

and removed

labels

2026-10-10 05:33:00 +00:00

sky_line referenced this issue

2026-10-10 12:52:36 +00:00

[Epic] Cluster LOD (meshlets + DAG, Nanite-style) for avatars, extensible to dense static mesh #642

sky_line commented