Virtual Shadow Maps

VSMs are designed to pair with Nanite and large dynamically lit worlds. If the target does not use Nanite, validate whether conventional shadow maps are a better fit.

Diagnose the two cost buckets

• Shadow Depths: rendering/updating caster depth pages.
• Shadow Projection under Lights: sampling/projecting shadows into the frame.

Use View Modes -> Virtual Shadow Map:

• Cached Page: green cached, red uncached, blue static-only cached;
• Shadow Casters: locate invalidators;
• Nanite Overdraw: locate costly caster coverage;
• Virtual Page/Clipmap-Mip: inspect page demand/resolution.

Disable visualization before timing. For counters, use trace.enable counters,vsm; CSV profiling can enable the VSM category.

Reduce invalidation

Largest invalidators commonly include moving lights, moving/added/removed casters, component render- state churn, WPO/PDO, skeletal deformation, large bounds, camera cuts, and rapid disocclusion.

Actions:

• stop needless light/primitive updates;
• keep deforming bounds tight;
• remove WPO/PDO or shadow casting at distance when visually safe;
• use material/mesh LODs that disable deformation in the distance;
• use Contact or Distance Field shadows where the content/target supports the trade;
• use Shadow Cache Invalidation Behavior overrides only when the asserted rigidity/static state is true; incorrect suppression produces undefined/stale shadows.

r.Shadow.Virtual.Cache 0 is a diagnostic to expose caching benefit, not a shipping optimization. Resolution LOD bias trades shadow detail for pages/work. Moving-light biases exist because moving lights cannot reuse cache like stationary ones.

When profiling individual VSM passes, r.RDG.AsyncCompute 0 can make timings more representative; never ship that profiling-only override.