Overseas media: Huawei is promoting its high-performance computing architecture based on the Ascend AI processor - Supernode 384, and hopes to gain broader support from developers.
This architecture is considered a competitor to Nvidia's NVL72 system of American semiconductor companies, aiming to alleviate bottlenecks in data centers. The president of Huawei's Ascend Computing Business said at the Kunlun Ascend Developer Conference held in Shenzhen that as the scale of parallel processing grows, cross-machine bandwidth in traditional server architectures has become a critical bottleneck for training, and future training needs will require innovative computing architectures like Supernode 384.
Huawei's super node architecture is the foundation of its previously announced CloudMatrix 384 system, which consists of 384 Ascend AI processors distributed across 12 compute cabinets and 4 bus cabinets, providing 300 petaflops of computing power and 48TB of high-bandwidth memory.
CloudMatrix 384 has been deployed in Huawei's data centers in Anhui, Inner Mongolia, and Guizhou, and is called the largest AI training platform in the industry. This architecture demonstrates Huawei's response strategy under U.S. technology restrictions, striving to break through the limits of AI system performance.
Original article: https://www.toutiao.com/article/1833252667760707/
Disclaimer: The article only represents the views of the author.