China’s LineShine tops TOP500 with 2 ExaFLOPS performance

China’s LineShine has taken the top spot in the June TOP500 ranking, which tracks 500 of the world’s most powerful supercomputers. The system delivers 2.198 ExaFLOPS on HPL, becoming the first to exceed 2 ExaFLOPS in sustained dual-precision FP64 workloads. Its theoretical peak is 2.736 ExaFLOPS, allowing it to sustain about 80% of maximum performance.

Unlike many modern data center systems optimized around 8-bit FP8 formats for AI workloads, LineShine reaches its FP64 mark with traditional CPUs and no GPUs or other accelerators. The result puts it ahead of U.S.-based El Capitan, Aurora, and Frontier while underscoring the continued role of FP64 computing in scientific HPC.

The machine comprises 13.79 million ARMv9 cores across 20,480 nodes in a non-uniform memory access configuration. Each node contains two LX2 CPUs, with 304 cores and 32 GB of HBM memory, while each CPU offloads to about 256 GB of system DDR memory. Each NUMA domain contains 38 cores running at 1.55 GHz and has 4 GB of HBM assigned to it, and each die is assigned 128 GB of DDR memory.