UB-Mesh routes NPU ports across rack, pod and superpod links
bookwormengr · x · 2026-07-22
This follow-up adds more of the paper’s architecture details for Huawei’s UB-Mesh training network.
- Ports are split across different mesh dimensions, so a chip can use links for intra-rack, inter-rack, and pod-level connectivity at the same time.
- The author contrasts this with Nvidia’s more fixed separation between NVLink and PCIe-based connectivity.
- The point is that a flat topology like CloudMatrix 384 or an Ascend 950 superpod needs many more switch chips and optical connectors, while UB-Mesh’s flexible port allocation can reduce that overhead.
More from Infra
- Qdrant plans a Bellevue event on AI memory and video retrieval — qdrant_engine · 2026-07-22
- Inference is turning into a tokenized commodity in a new onchain AI market map — 0xJeff · 2026-07-22
- Google DeepMind commits $40M in AI tokens and cloud credits for U.S. Energy Department research push — GoogleDeepMind · 2026-07-22
- Jensen Huang says Chinese open models should be used, not feared — hyhieu226 · 2026-07-22
- AWS may be entering a longer brain-drain phase after Dave Brown’s 19-year exit — DavidLinthicum · 2026-07-22
- Glow exits stealth with $180M at a $1.2B valuation to secure AI agents on endpoints — Justgototheeffinmoon · 2026-07-22