MetaX 曦云C500X 光链超节点 (Optical Supernode)

64x C500X cluster

Overview

—

Peak Dense

—

Aggregate Memory

—

Aggregate Bandwidth

—

Aggregate Power

Configuration

Accelerator: C500X
Count: 64
Memory per GPU: 64 GB
Bandwidth per GPU: —

Power & Efficiency

Aggregate Power: —
TDP per GPU: 350 W
Efficiency: —
Aggregate Bandwidth: —

System Details

Vendor: MetaX

Family: 曦云

Architecture: —

Form Factor: cluster

Release Year: —

Process Node: —

Interconnect

DragonFly topology over hybrid optical and electrical MetaXLink, spanning up to eight chassis

Notes

MetaX's optically interconnected supernode. Hybrid optical and electrical MetaXLink implements a DragonFly topology across up to eight chassis for a maximum of 64 曦云C500X cards, and because the building block is a conventional PCIe server the node fits air-cooled, low-power racks rather than requiring a liquid-cooled cabinet. MetaX presents this as the cost-effective route to cross-cabinet scale-up, against the 3D Mesh node's electrical links and the Shanghai Cube's density. The 64 cards shown here are the maximum topology MetaX states rather than a fixed configuration. No power, memory total or throughput figure is published.