China's CPU-Only LineShine Tops the TOP500, First Chinese Entry in 9 Years

At ISC 2026 in Hamburg, the 67th TOP500 list crowned a new leader: LineShine, a Shenzhen-based system and the first Chinese submission in nine years. Unusually, it’s CPU-only, built on a domestic Armv9 chip called the LX2 — a 304-core, 1.55 GHz package with 228 MB of L2 cache, 32 GB of on-package high-bandwidth memory (likely an indigenous design rather than standard HBM), and 256 GB of DDR5 spillover per chip. Scaled across roughly 22,000 nodes and 13 million cores, LineShine hits 2.198 sustained exaflops (Rmax) while drawing 42.22 MW. More notable than the LINPACK score: it also tops the HPCG benchmark at 22 petaflops, beating El Capitan, signaling a genuinely HPC-focused design rather than a benchmark special.

The rest of the list saw smaller moves. Eni’s HPC7 in Italy debuts at number 6 — essentially a 30%-scale El Capitan on the same HPE Cray EX4000 platform with AMD MI300A APUs — giving Italy more listed compute than any other European country. Fugaku slipped to ninth overall but remains third on HPCG six years after launch. The Green500 efficiency top 10 went unchanged for the first time ever, though field-wide efficiency improved as older systems retired.

The author frames LineShine as a geopolitical marker: China has had exascale machines for years but never put one on the list, and it may keep others (Sunway Oceanlight, CNIS) off it. The submission could pressure the US DOE toward more funding for systems like Discovery. The piece also raises why hyperscale AI clusters — xAI’s Colossus 2 with 550,000+ Blackwell GPUs — never run HPL or appear on the TOP500. Finally, governance of the list is moving from the ISC Group to ACM SIGHPC, which will assign DOIs to future lists.