Group 1 - Huawei unveiled its AI computing capabilities at the Huawei Connect Conference, showcasing a comprehensive AI computing architecture that includes new Ascend chips, supernodes, and a new interconnect architecture called Lingqu [1][2] - The company emphasized that computing power is crucial for artificial intelligence and is a key component for China's AI development, aiming to create a "supernode + cluster" computing solution to meet the growing demand for computing power [1][3] - The Ascend chip series will see rapid iterations, with the Ascend 950 series set to launch in Q1 2026, followed by the Ascend 960 in 2027 and the Ascend 970 in 2028, with a goal of doubling computing power annually [2][3] Group 2 - Huawei's Kunpeng ecosystem is rapidly evolving, with the Kunpeng 950 expected to launch in Q4 2026 and the Kunpeng 960 in Q1 2028, positioning itself against Intel and AMD [3][4] - The company has developed a complete AI hardware and software system, moving beyond single-card solutions to clusters and supernodes, which are essential for AI infrastructure [4][5] - Huawei introduced the Atlas 950 SuperPoD and Atlas 960 SuperPoD supernodes, which support 8192 and 15488 Ascend cards respectively, claiming to be the world's strongest computing supernodes [5][6] Group 3 - Huawei has launched the world's first general-purpose computing supernode, TaiShan 950 SuperPoD, which aims to replace various large and small machines, showcasing its innovation in interconnect technology with the Lingqu protocol [6][7] - The company is focusing on creating a sustainable computing industry based on available chip manufacturing processes, emphasizing the need for system-level computing rather than just single-processor solutions [7][8] - Huawei's approach to AI computing is characterized by systematic innovation and a focus on building a robust computing ecosystem, aiming to provide new computing options globally [3][6]
从超节点到集群 华为亮出AI算力全家桶
2 1 Shi Ji Jing Ji Bao Dao·2025-09-18 23:46