ZOZO open-sources GPU contact solver for 180M-point simulation
Original: ZOZO Open-Sources GPU Contact Solver Handling 180M Contacts View original →
A production contact solver moves into the open
ZOZO’s internal physics tooling is now public, giving simulation researchers and graphics engineers a new way to test cloth, rope, and soft-body contact at unusual scale. In the source tweet, Gorden Sun highlighted that the solver can handle “over 180 million contact points” in a single scene. The original tweet is available here.
The linked repository, ppf-contact-solver, describes the project as a contact solver for physics-based simulations involving shells, solids, and rods. The README says it was built by ZOZO, Japan’s largest fashion e-commerce company, and emphasizes penetration-free contact resolution, strict strain limits for triangles, GPU execution, Python APIs, Docker deployment, and a Blender add-on. It is also permissively licensed under Apache 2.0, which matters for studios and product teams evaluating whether the code can enter commercial pipelines.
The broader AI relevance is that generative workflows increasingly need reliable physical feedback. Avatar tools, virtual try-on systems, robotics simulators, and 3D asset agents can create plausible shapes quickly, but their output often breaks down when cloth, soft bodies, or dense contact interactions need to obey constraints. A solver that exposes documented APIs and even mentions MCP support points toward agent-driven simulation loops where an LLM can set up or run experiments instead of only writing helper scripts.
The next signal to watch is adoption outside ZOZO’s examples. The project requires modern NVIDIA hardware and CUDA 12.8 or newer for serious workloads, and the maintainers warn that it is built for offline rather than real-time use. Community Blender demos, benchmark reproductions, and integrations with existing digital fashion or robotics stacks will show whether the 180M-contact claim becomes practical infrastructure or remains a specialized research-grade tool.
Related Articles
The Google DeepMind spinout closed a $2.1 billion Series B led by Thrive Capital to scale its AI drug design engine (IsoDDE), building on Nobel Prize-winning AlphaFold technology.
Penn researchers demonstrated optical signal switching using exciton-polaritons — hybrid light-matter particles — consuming roughly 4 femtojoules per operation. The breakthrough removes a key obstacle to fully optical AI inference hardware.
Google DeepMind unveiled Gemini for Science at I/O 2026, a suite of experimental AI tools designed to help scientists explore hypotheses, validate work at scale, and analyze scientific literature.
Comments (0)
No comments yet. Be the first to comment!