Foreign Media: China Launches the Largest Domestic AI Computing Cluster for Trial Operation, Promoting Infrastructure Construction
On February 6, the National Supercomputing Internet of China (SCNet) launched a new node test in Zhengzhou, Henan, driven by the country's largest domestic AI computing cluster.
This node consists of three scaleX platforms provided by SUGON, supporting the deployment of more than 30,000 AI accelerator cards. Although SCNet did not disclose the total computing power, based on the performance of a single scaleX platform, the combined computing power of the three platforms can reach 15 eflops (10^24 floating-point operations per second).
As a reference, the world's most powerful supercomputer currently is HPE's El Capitan, with a peak performance of 1.8 eflops (according to the latest TOP500 ranking). This means that the theoretical computing power of this Chinese node is about 8.3 times that of the current world champion.
This node is designed for large-scale AI scenarios, including training of trillion-parameter models and high-throughput inference. This move marks a significant advancement in China's independent development of AI computing infrastructure.
Original article: toutiao.com/article/1856431613153544/
Statement: The article represents the views of the author.