China trials its largest locally made AI computing resource in major infrastructure push
The node is designed for large-scale AI scenarios, including trillion-parameter model training and high-throughput inference
China’s state-backed SuperComputing Network (SCNet), a major computing resource provider, has deployed a new node for beta testing, powered by the country’s largest domestically made artificial intelligence computing cluster.
The node, which began trial runs on Thursday in Zhengzhou, central Henan province, consists of three scaleX platforms from Chinese supercomputer developer Sugon, which support the deployment of more than 30,000 AI acceleration cards, according to statements from both SCNet and Sugon.
Sugon is publicly listed in Shanghai as Dawning Information Industry.
SCNet did not disclose the total computing power of the node, but based on the performance of a single Sugon scaleX, the three platforms could deliver a combined computing power of 15 eflops.
An eflop, short for exa floating-point operations per second, measures the performance of supercomputers and high-performance AI hardware that can calculate at least one quintillion floating-point operations per second.
Currently, the world’s most powerful supercomputer is El Capitan from Hewlett Packard Enterprise, which achieved 1.8 eflops at its maximum performance, according to the latest list from TOP500, a project ranking high-performance computing systems.