Performance and Big Data

Intermact rendering optimizes for "many objects / large point clouds / heavy resampling" along three paths (design.md §15.2): GPU instancing, Float32Array buffer channels, Worker off-main-thread sampling. Core stays DOM/GL agnostic.

GPU instancing

instanceField(object, transforms) expresses massive repetition with single base geometry + per-instance transforms. It both:

• Bakes aggregate geometry into stroke/fill traits — headless sampling, SVG, picking, bounds all correct;
• Carries InstancedTrait; GPU renderer draws base geometry once per instance via true InstancedMesh (not giant aggregate buffer).

Renderers without instancing path fall back to baked group geometry (pixel-identical, no GPU acceleration). Instanced views disable frustum culling (or extend bounds) to avoid incorrect culling, and handle morph geometryOverride and trait change redispatch.

import { circle, instanceField } from "@intermact/core";
const dots = instanceField(
  circle({ radius: 0.02 }),
  Array.from({ length: 10000 }, () => ({ position: [Math.random() * 8 - 4, Math.random() * 6 - 3] })),
);

Large point clouds: Float32Array channel

pointCloud3D consumes Float32Array position buffers directly; renderer uploads via that channel, no per-frame geometry rebuild. Create reveals by clipping draw range, avoiding per-frame churn — 60k points seek smoothly.

Worker sampling

Heavy polygonization (e.g. marching-cubes) can run off main thread: @intermact/render-three provides worker protocol/kernel/client glue; core stays pure. Main thread and Worker paths produce identical geometry.

Geometry memoization and budgets

• Sampling/triangulation results memoized by content (geometry/memoize); same input reuses buffers.
• Timeline seek is pure function: seek on 2000-track interleaved timeline remains correct and fast (see perf/perf-budget tests).

Related examples

• perf/instanced-10k — instanceField GPU instancing 10k objects
• perf/large-pointcloud — Float32Array channel large point cloud
• perf/worker-sampling — Worker off-main-thread marching-cubes sampling

Full list in example index.