SIGGRAPH cloth paper makes CPUs competitive again
This SIGGRAPH 2025 (ACM TOG 44(4)) paper introduces a multicore CPU cloth simulator that partitions garments into domains and couples boundaries efficiently inside a projective-dynamics framework. The authors report at least an order-of-magnitude speedup over prior CPU methods, with performance that is often comparable to strong GPU cloth solvers on their showcased workloads.
Smart decomposition, not raw hardware brute force, is the key idea here, and it meaningfully shifts what “CPU-only” cloth simulation can do.
- –Domain-level parallelization targets multicore CPU architecture directly, which helps close the usual GPU gap.
- –The reported gains come with contact-rich demos (self-collision, tangling, layered tablecloths), not just easy scenes.
- –If these results generalize, studios can reserve GPUs for rendering/ML while still running high-quality cloth sims on CPUs.
- –The “CPU can match or beat GPU in shown cases” framing is exciting, but practical impact will still depend on scene mix and implementation constraints.
DISCOVERED
71d ago
2026-03-17
PUBLISHED
71d ago
2026-03-17
RELEVANCE
AUTHOR
Two Minute Papers