Huawei Unveils AI Supernode to Challenge Nvidia’s Dominance
华为发布新型AI架构，挑战英伟达霸主地位
Huawei has introduced the CloudMatrix 384 Supernode, claiming it can rival Nvidia’s NVL72 system in tackling computing power bottlenecks in AI data centers. The device is said to deliver 300 petaflops, surpassing Nvidia’s 180. While Nvidia’s NVL72 uses 72 GPUs linked via NVLink to support ultra-fast inference for large AI models, Huawei’s system is already running in its Wuhu data centers. Though the company hasn’t officially commented, the launch signals Huawei’s progress toward computing self-sufficiency, despite ongoing U.S. sanctions.
华为推出了名为 CloudMatrix 384 Supernode 的新型AI架构，声称可与英伟达的 NVL72系统 抗衡，解决AI数据中心的算力瓶颈问题。该设备计算能力达 300 petaflops，超过英伟达的180。英伟达的NVL72通过 72个GPU及NVLink高速互联技术 支持大模型的超高速推理，而华为的系统已在其位于安徽芜湖的数据中心投入使用。尽管华为尚未正式回应，但这一发布被视为其在 突破美方技术封锁、实现算力自给 上迈出的关键一步。
