A Domestic Accelerator Enters the Top-10 List
On 15 November, at the 8th China Supercomputing Conference (ChinaSC 2026) in Beijing, the National Supercomputing Centre in Shenzhen’s “Lingsheng” took first place in the 2026 China HPC TOP100 with 2.198 EFlops of measured Linpack double-precision performance — roughly 4.5 times last year’s leader and a follow-up to its June 2026 world TOP500 crown at 2.19 EFlops, 22% clear of El Capitan. The system runs an all-CPU “online acceleration” architecture across about 90 cabinets, 22,680 nodes and more than 13 million cores with self-developed LX2 CPUs and domestic HBM, sustaining over 80% parallel scaling efficiency. The signal sits at number 10: a system built on domestic MetaX accelerator cards appears on a national TOP100 for the first time, the same supply chain thread as the chip-funding rotation and Xixuan Chuangzhi covered in earlier rounds.
