China’s LineShine Supercomputer Tops TOP500 with Pure‑CPU Architecture, Outpacing US Systems

Core Performance Data

LineShine achieved a sustained double‑precision performance of 2.198 EFlops in the standard HPL benchmark, making it the first E‑class supercomputer to break the 2 EFlops barrier. The runner‑up, the US El Capitan at Lawrence Livermore, recorded 1.809 EFlops, meaning LineShine is about 21% faster.

Beyond HPL, LineShine also ranked first in the HPCG benchmark and fourth in the mixed‑precision HPL‑MxP test, demonstrating balanced strength across multiple workloads.

Technical Route: Fully Domestic Chipset

Unlike most contemporary E‑class systems that rely on GPU accelerators, LineShine uses a pure‑CPU architecture built on the self‑developed LX2 processor. The system comprises 13.79 million CPU cores, each 1.55 GHz, with 304 cores per chip, based on a heavily customized ARMv9 ISA and integrated matrix‑compute units to serve both scientific and AI workloads.

The supporting stack includes the home‑grown “LingQi” high‑speed interconnect, domestically produced HBM memory, and the Kylin operating system, achieving a total power consumption of roughly 42.2 MW while maintaining competitive energy efficiency.

Overall Ranking Landscape

The TOP500 list remains dominated by the United States and China. The United States holds the most entries, occupying positions 2–4, while China has two machines in the top ten: LineShine at #1 and Sunway TaihuLight at #9.

Globally, the number of E‑class supercomputers has risen sharply, but LineShine is the only system to exceed the 2 EFlops sustained performance threshold.

Application Scenarios and Industry Value

LineShine will provide computing resources for nationwide research and industry, covering climate modeling, aerospace simulation, new‑energy material research, biomedical molecular simulation, and large‑model AI training.

Jack Dongarra, co‑founder of the TOP500 list and Turing Award laureate, highlighted that LineShine proves a pure‑CPU system can reach E‑class performance, offering a viable path for nations facing export restrictions on high‑end GPUs.

Conclusion

LineShine’s ascent to the TOP of the TOP500 list marks the culmination of years of domestic processor development, demonstrating that China can now build the world’s most powerful supercomputers without foreign accelerator chips and opening a new direction for CPU‑centric AI‑integrated high‑performance computing.