Huawei CM384

Huawei CM384 is the domestic [[AIAcceleratorSupernode|AI accelerator supernode]] case highlighted in 国产 AI 算力能凭「超节点」弯道超车吗？｜WAIC 深度观察 S10E23. The source presents it as China’s most widely discussed supernode comparison point against [[NvidiaGB200NVL72|Nvidia GB200 NVL72]].

The episode’s core tradeoff is scale versus efficiency. CM384 is described as using many more Huawei accelerators to reach higher aggregate BF16 compute than NVL72, while also carrying much higher cited power consumption. That makes CM384 a useful case for Domestic AI Chip Catch-Up: system-level design can offset weaker per-chip performance, but the result still has to pass power, cooling, software, and customer-order tests.

Connections

• Huawei and HiSilicon — company and chip-design context.
• AI Accelerator Supernode and Scale Up AI Interconnect — system category and interconnect logic.
• Nvidia GB200 NVL72 — comparison case in the source.
• Data Center Power Bottleneck, Data Center Thermal Management, and Domestic AI Chip Order Validation — constraints that decide whether the system is usable at scale.