华为云CloudMatrix384平均单卡吞吐达H20的3-4倍

Core Insights - Huawei has significantly upgraded its Ascend AI cloud services, enhancing its CloudMatrix platform to support large-scale AI computing needs [1] Group 1: Service Upgrades - The CloudMatrix cloud supernode specifications have been greatly enhanced, enabling ultra-large clusters and providing substantial computing power [1] - Huawei Cloud has introduced an elastic memory storage service (EMS) to improve the performance of large models [1] - The CloudMatrix 384 AI Token service has been fully launched to meet industry AI computing demands, with the innovative supernode architecture greatly enhancing system inference performance, achieving an average single-card inference performance 3-4 times that of H20 [1] Group 2: Infrastructure and Efficiency - Huawei Cloud is deploying large-scale, fully liquid-cooled AI data centers to reduce energy consumption [1] - The company offers a full-stack dedicated AI cloud service to ensure data security [1] Group 3: Market Growth - The number of customers for Ascend AI cloud services has increased from 321 last year to 1,805 this year [1] - The total AI computing power scale has grown by 268% year-on-year, providing strong momentum for intelligent upgrades across various industries [1]