Arm-Based LineShine Supercomputer Takes Top500 Crown, Surpasses 2 Exaflops

The June 2026 Top500 supercomputer list, released Tuesday at the ISC 2026 conference in Hamburg, Germany, has a new number one. China's LineShine system, installed at the National Supercomputing Centre in Shenzhen (NSCS), achieved 2.198 exaflops on the High Performance Linpack benchmark, displacing the U.S. Department of Energy's El Capitan.

LineShine is the first system on the Top500 to surpass 2 exaflops of sustained performance. It is roughly 22 percent faster than El Capitan, which held the top spot since November 2024. The system consumes 42 megawatts of power, yielding about 52 gigaflops per watt, slightly below El Capitan's 60 gigaflops per watt.

CPU-Only Design with 304-Core Arm LX2 Processors

Unlike the previous top systems Frontier and El Capitan, which rely on GPUs driven by CPUs, LineShine is a purely CPU-based machine. Its fundamental building block is the LX2 processor, believed to be designed by Huawei. Each LX2 chip contains two compute dies, each with 152 Armv9-architecture cores supporting SVE and SME vector and matrix extensions. A single die is divided into four NUMA domains of 38 cores, each with 4GB of HBM and 32GB of DDR5. The full CPU has 304 cores, 32GB of HBM, and 256GB of DDR5, consuming 690 watts. Integrated 800Gbps NICs are also on chip.

• Each node contains two LX2 CPUs (608 cores). Eight nodes form a blade (16 CPUs). Sixteen blades make a compute frame (256 CPUs). Two frames go into a cabinet (512 CPUs). The full system has roughly 89 to 90 cabinets, totaling about 13.79 million cores.
• The system uses a LingQi 1.6Tbps network with a four-layer fat-tree topology and one optical layer.
• LineShine achieves 80 percent of its 2.736 exaflops theoretical peak, a higher efficiency than U.S. exascale systems, which range from 50 to 65 percent.
• The system is liquid-cooled to manage the high thermal design power of the LX2 CPUs.

Implications for HPC and Arm Architecture

LineShine's debut marks the first time a non-Lenovo Chinese system has appeared on the Top500 in several years. It also represents a major win for Arm computing in high-performance computing, demonstrating that CPU-only designs can compete with GPU-accelerated systems at the exascale level. The LX2's integration of HBM and high-speed networking on chip points to a trend toward more tightly coupled, power-dense compute nodes.

Looking ahead, the Top500 list is expected to see more Arm-based entries as Chinese and other vendors develop similar processors. The power efficiency gap between LineShine and GPU-based systems may narrow as Arm designs mature. The next Top500 update, scheduled for November 2026, will show whether LineShine can hold its lead or if new systems from the U.S. or Japan will challenge it.