Overview
- LineShine debuted at No.1 on the TOP500 after submitting results on Tuesday, June 23, 2026, registering 2.198 exaflops on the double‑precision HPL benchmark.
- The system runs on a domestic LX2 CPU design with about 13.79 million cores, a proprietary LingQi interconnect, Kylin OS, and consumes roughly 42.2 megawatts of power.
- LineShine is the first TOP500 entry to sustain over 2 FP64 exaflops using only CPUs, a feat that highlights Chinese progress in an independent high‑performance computing supply chain.
- On AI‑relevant mixed‑precision tests (HPL‑MxP) LineShine lags leading GPU‑accelerated machines, and its GFLOPS-per‑watt efficiency is lower than top U.S. systems, limiting its appeal for many modern AI training tasks.
- The TOP500 ranking measures HPL double‑precision throughput for scientific simulation, not overall AI capability, and many hyperscaler AI clusters do not appear on the public list so the full global picture remains incomplete.