crowdbricks: open-source batching hits 10k motion generations/sec on one RTX 4090, 26x faster
TinfoilTricorn · x · 2026-10-28
- NVIDIA's open-sourced MotionBricks generates real-time motion for game characters and humanoids: given recent poses and target poses, it fills in the movement at 2.6 ms per request on an RTX 4090
- But it serves one character at a time, so crowds get slow as every character waits its turn
- The open-source crowdbricks project rebuilds the same ONNX model as a TensorRT plan with a batch dimension, generating motion for hundreds of characters in a single call
- Result: 10,000 generations per second on one RTX 4090, 26x faster than one-at-a-time, still realtime — enabling insanely dynamic in-game crowds
- MotionBricks was trained on 350K+ motion clips, with SOMA weights for human characters and Unitree G1 weights for humanoids; code Apache 2.0, weights under NVIDIA Open Model Agreement
More from Embodied
- MOSS-Image character LoRA trained on RX 7900 XT 20GB in 3.5 hours on native Windows — iBorgSimmer · 2026-10-28
- SuperMimic runs on musculoskeletal humanoids as robotics iteration cycles shrink to hours — micoolcho · 2026-10-28
- Generalist Contact Planner + Tracker + Neural Mapping Push Humanoid Mobility Forward — ChongZzZhang · 2026-10-28
- Machenta CEO bets on robots that learn tasks from a single demo — Ed__Johns · 2026-10-28
- Microduck team spends a month in Shenzhen to finalize mass production of its humanoid robot — victormustar · 2026-10-28
- Tesla Optimus lines target 1M robots/year, public sales eyed for late 2027 — anthara_ai · 2026-10-28