UB-Mesh: A Hierarchically Localized nD-FullMesh Data Center Network Architecture

The scaling of large-scale language models (LLMs) demands unprecedented computational power and bandwidth. We present unified bus (UB)-Mesh, an innovative artificial intelligence data center network architecture that enhances scalability, performance, and cost-efficiency through a hierarchical nD-FullMesh topology. Unlike traditional symmetrical designs, UB-Mesh optimizes LLM training by prioritizing localized data movement and minimizing switch usage. The architecture features UB-Mesh-Pod, a physical implementation of 4D-FullMesh using custom hardware, including neural network processing units, CPUs, low/high-radix switches, and network interface cards, interconnected via our UB technology for dynamic resource allocation. For network optimization, we introduce all-path routing to efficiently manage data traffic. Combined with topology-aware performance tuning and robust reliability mechanisms, like 64 + 1 backup, UB-Mesh achieves 2.04× better cost-efficiency and 7.2% higher availability than Clos networks. These innovations address the critical challenges of building practical, high-performance artificial intelligence infrastructure at scale.