Overview
GPU rigid bodies, fluids, and cloth as one operator suite — eight operators, one solver clock.
Matter is a simulation suite, not a collection of solvers. Rigid bodies, fluids, cloth, and constraints all step on one clock inside one CUDA context — which is what lets a fluid push a box that tears a cloth, in real time, without a frame of latency between them.
Matter runs on any CUDA 7.5+ GPU — RT cores not required. A GTX 16-series laptop runs the small examples; see Requirements.
The suite
Eight operators
One solver, three body types, two force fields, an emitter, and a readback CHOP — the whole surface.
Two-way coupling
Solvers exchange impulses inside a substep, not across frames — the feature the suite is built around.
Design rules
- Determinism switch.
Deterministicon the solver trades ~15% speed for bit-identical replays — same inputs, same frames, every run. - SOPs in, SOPs/TOPs out. Colliders and emitters are plain geometry; results come back as point clouds, meshes, or velocity fields.
- Budgeted stepping. Like Photon OVRTX, the solver takes a millisecond budget and reduces substeps before it ever stalls your frame.
FAQ
No — Matter's solvers are written for the GPU-resident, two-way-coupled case. A wrapper around a CPU engine can't share substep impulses with a GPU fluid without copying state across the bus every step.
Placeholder targets on a 4080: ~20k active rigid bodies, ~2M fluid particles, or ~500k cloth vertices — budget-shared when mixed. Real numbers land with the benchmark project.
The readback CHOP/SOP pair feeds TouchDesigner's native caching (Cache SOP, File Out). A dedicated bake operator is on the roadmap.