China's LineShine supercomputer, based on Arm CPUs, has retaken the #1 position on the June 2026 Top500 list with 2.2 exaFLOPS performance—China's first #1 ranking since 2017.
In a significant milestone for Arm computing, China's LineShine supercomputer has claimed the number one position on the latest Top500 supercomputer list. Developed by the National Supercomputing Centre in Shenzhen (NSCS) and the first non-Lenovo Chinese system submitted in considerable time, LineShine is the first to publicly exceed 2 ExaFLOPS on Linpack. With a measured performance of just under 2.2 ExaFLOPS, it delivers approximately 22 percent greater performance than El Capitan, the classified U.S. system that previously held the top ranking.
The most striking characteristic of LineShine is its architecture. Unlike the previous top systems, El Capitan and Frontier, which rely on GPUs paired with CPUs, LineShine is a pure CPU system with no GPUs. The foundation of LineShine is the LX2 CPU, believed to be designed by Huawei. Each LX2 CPU comprises two compute dies with onboard HBM and 152 Armv9-architecture cores per die, both equipped with SVE and SME vector and matrix acceleration blocks. This enables GPU-like performance without traditional graphics processors.
Each LX2 die is organized into four NUMA domains containing 38 cores each, with 4GB of HBM and 32GB of DDR5 per domain. The configuration can operate as either a flat topology combining HBM and DDR capacity or as a cache topology with HBM serving as a cache for DDR. Across the entire chip, there are 304 cores, 32GB of HBM, and 256GB of DDR5. To achieve GPU-comparable performance, the LX2 processor draws 690 watts of power, according to a presentation at the International Forum for HPC & AI Co-driven Innovation. The chip also integrates 800Gbps NICs on-die, a level of integration rarely seen since the era of Intel Xeon processors with OmniPath.
The system architecture scales these building blocks into a hierarchical structure. Two 304-core CPUs form a node. Eight nodes comprise a blade containing sixteen CPUs. Sixteen blades create a compute frame holding 256 CPUs. Two compute frames form a cabinet with 512 CPUs. With approximately 89 to 90 compute cabinets total, the system contains roughly 13.79 million cores, with each cabinet housing over 155,000 cores. A LingQi interconnect operating at 1.6 Terabits per second connects to each node, employing a four-layer fat-tree topology with an optical layer.
LineShine achieved 2.2 ExaFLOPS in measured performance versus a theoretical maximum of 2.74 ExaFLOPS, representing 80 percent of peak performance. This efficiency notably exceeds that of competing GPU-based exascale systems, which operate between 50 and 65 percent of theoretical peak. However, the system is power-intensive, consuming 42 megawatts according to the official Top500 submission, yielding approximately 52 gigaflops per watt—slightly below other exascale systems such as El Capitan at 60 gigaflops per watt, a difference attributable to the lower compute density of CPUs compared to GPUs. The system employs liquid cooling to manage its high thermal output.
LineShine's arrival marks a significant shift in the Top500 rankings. The threshold to enter the top five now exceeds 1 ExaFLOPS, with Germany's JUPITER Booster—Europe's sole exascale system—now occupying the fifth position. For the first time in four years, an AMD-powered supercomputer no longer leads the list. El Capitan, powered by AMD MI300A processors, now ranks second, while Frontier, the world's first exascale system based on EPYC and MI250X processors, drops to third. A new entry, the Italian Eni S.p.A. HPC7, claims the sixth position with 571 petaflops of performance using AMD MI300A technology in a smaller version of the El Capitan architecture, consuming only 8.7 megawatts compared to El Capitan's higher power draw.
AMD now provides the processors for four of the top ten systems on the list. NVIDIA maintains three Hopper-architecture systems at positions five, seven, and ten. Aurora, an Intel GPU-powered system, remains at fourth place and is the sole Intel GPU system on the list, though Intel CPUs also support the GPU-based Eagle system at seventh. LineShine joins Fugaku, Fujitsu's previous CPU-only supercomputer based on Armv8 architecture, which now holds the ninth position following these latest additions.
It is worth noting that many large-scale computing systems worldwide are not included in the Top500 Linpack benchmarks. Additionally, the industry is witnessing a trend toward AI hardware designs that eliminate FP64 support in favor of enhanced mixed-precision performance capabilities.