LineShine Claims TOP500 #1 at 2.198 Exaflops Rmax with 13 Million Arm Cores

The 67th TOP500 list at ISC 2026 records the first Chinese system at number one since 2017. LineShine deploys 22,000 nodes containing 13 million LX2 cores organized as dual-socket blades with 800 Gbps interconnects per node. The machine records 2.198 exaflops Rmax against 2.735 exaflops Rpeak while drawing 42.22 MW, and simultaneously leads the HPCG ranking at 22.004 petaflops, surpassing El Capitan.

LX2 silicon integrates 304 active Armv9 cores per package at 1.55 GHz with 32 GB on-package memory delivering 4 TB/s bandwidth plus 256 GB DDR5 per socket. Measured efficiency reaches 52.07 GFLOPS/W on LINPACK. This architecture avoids GPU accelerators yet achieves both peak LINPACK and leading HPCG scores, indicating sustained memory and interconnect balance rather than benchmark-specific tuning.

The result shifts measured global FP64 capacity leadership to a fully domestic Chinese design for the first time in nine years. Italy’s HPC7 at 571.5 petaflops now places Europe’s largest listed system outside Germany. Because the submission uses only Arm CPUs at scale, it demonstrates an alternative path to exascale that does not rely on US-sourced accelerators, directly affecting future national AI training throughput measured in FP64-equivalent operations.

No Green500 turnover occurred, yet aggregate list efficiency rose through retirement of older nodes. LineShine’s dual dominance on LINPACK and HPCG suggests production workloads will see measurable gains in sustained throughput compared with prior Chinese entries.