China’s LineShine supercomputer claimed the top slot on the June TOP500 list, delivering a sustained 2.198 ExaFLOPS on the HPL benchmark. That performance makes it the first machine to break the 2 Exa barrier in dual‑precision FP64 workloads, and it achieves roughly 80 % of its 2.736 ExaFLOPS peak. Unlike most exascale systems, the run relies solely on CPUs, with no GPUs or accelerators.

The machine stitches together 13.79 million ARMv9 cores across 20,480 nodes in a NUMA layout. Each node houses two LX2 CPUs, 304 cores total, 32 GB of HBM and about 256 GB of DDR memory, likely DDR5. Running at 1.55 GHz, the system draws 42.2 MW and records 52.07 GFLOPS per watt, the only CPU‑only exascale entry in the current TOP500. The configuration also supports high‑speed interconnects for massive parallelism.

By eclipsing U.S. machines such as El Capitan, Aurora and Frontier, LineShine signals China’s growing dominance in scientific HPC, where FP64 remains the benchmark for climate, physics and materials simulations. The CPU‑only architecture revives interest in ARM‑based designs, offering a path that avoids the costly GPU ecosystems dominating AI workloads. It shows ARM clusters can rival GPUs. Its verified performance cements the nation’s first TOP500 victory.