China's Lingsheng supercomputer ranks among the world's top 500, with sustained performance setting a world record

[ Gearbest Technology News]On June 23, at the ISC2026 conference in Hamburg, Germany, the latest global supercomputing TOP500 list was officially released. The new generation “Lingsheng” supercomputer released by the National Supercomputing Center in Shenzhen ranks first with a sustained double-precision floating-point performance of 2.19EFlops, becoming the world's first supercomputing system with a sustained performance exceeding 2EFlops. This is the first time in nine years that Chinese supercomputing has topped the list again.

“Lingsheng” supercomputer

Lu Yutong, chief designer of the “Lingsheng” supercomputing system, director of the National Supercomputing Center in Shenzhen, and professor of Sun Yat-sen University, delivered a special speech at the award ceremony, explaining in detail the core design concepts and technical routes of the system. She said that the development of supercomputing has entered a new stage of super-intelligence integration. The “Lingsheng” system pioneered the Online Acceleration full CPU architecture, breaking the barriers of traditional CPU and GPU heterogeneous architectures. It embeds AI matrix acceleration units inside the CPU, returns to the essence of computing acceleration, and achieves efficient collaboration of multiple computing modes such as supercomputing and intelligent computing.

Professor Jack Dongarra, winner of the Turing Award and academician of both the US Academy of Sciences and the US Academy of Sciences, spoke highly of this system, saying that China's “Lingsheng” has allowed the world to see the light of hope leading to the new system architecture of AI4Science.

Judging from the technical details, “Lingsheng” is positioned as a domestic full-stack integrated infrastructure for scientific and engineering intelligent computing, achieving independent innovation at multiple levels. At the chip level, the self-developed LX2 CPU introduces multi-precision and matrix acceleration capabilities, achieving a deep integration of on-chip supercomputing power and intelligent computing power, and integrates the first domestic on-chip memory. The memory bandwidth is increased by 10 times compared to traditional CPUs. At the network level, the self-designed Lingqi high-speed interconnection network can support ultra-large-scale networking with 2 million ports and 100,000 nodes. The storage layer adopts a layered architecture, taking into account high-performance operating areas and capacity-based data areas, and can be expanded to exascale scale. At the system level, a large-scale super-intelligent fusion framework has been built, and the underlying hardware capabilities are released to the application layer in an available, programmable, and optimizable manner through self-developed full-stack software. In terms of energy saving, it is the first 100% fully liquid-cooled computer cabinet with an energy efficiency ratio of 51GFlops/W.

In terms of application verification, since its deployment, “Lingsheng” has supported applications in many fields such as atmosphere and ocean, engineering simulation, materials science, drug discovery, brain science, scientific AI, and large model reasoning. The average expansion efficiency is 84.4% in a large-scale parallel environment, achieving efficient and scalable operation of the entire system with more than 10 million cores.