BufferAttribute vs InterleavedBuffer: performance trade‑offs on diverse GPUs

Goal

Identify the attribute layout that yields the lowest GPU‑side latency when rendering large, frequently updated geometries across a range of desktop and mobile GPUs.

Constraints

• Three.js BufferGeometry supports both separate BufferAttribute objects and a single InterleavedBuffer with multiple InterleavedBufferAttribute views.
• Driver hints (StaticDraw, DynamicDraw) affect upload strategy but do not dictate memory layout.
• Cache locality, vertex fetch bandwidth, and memory fragmentation differ between integrated and discrete GPUs.

Open questions

1. Does a single large InterleavedBuffer consistently outperform multiple small BufferAttributes on modern mobile GPUs?
2. At what vertex count does the overhead of separate attribute bindings outweigh the flexibility of independent updates?
3. Are there measurable differences when the usage hint is set to DynamicDraw versus StaticDraw for each layout?