China’s LineShine remains world’s fastest supercomputer

Network World ·

China’s LineShine remains world’s fastest supercomputer

China’s LineShine again topped the list of the world’s fastest supercomputers in the 68 th edition of the TOP500 announced this week at SC26 in Chicago. All 10 leaders held their June positions, and just 27 new systems joined the list, which is the fewest newcomers in its history. LineShine, installed at the National Supercomputing Centre in Shenzhen, retained its lead with 2.198 exaflops on the High Performance Linpack (HPL) benchmark. The CPU-only system runs on custom LX2 processors connected by the proprietary LingQi interconnect. El Capitan held second place at 1.809 exaflops, followed by Frontier , Aurora, and Germany’s JUPITER Booster. Those five systems are the only ones on the November list to reach an exaflop on HPL. The bar for making the TOP500 rose even with record-low turnover. The entry threshold rose 7.3% from June to 2.857 petaflops. Across all 500 systems, combined HPL performance climbed 4.4% to 19.57 exaflops, and 291 systems, or 58.2% of the list, used accelerators. TOP500 organizers suggested that more new hardware is going into large AI clusters. They cited Epoch AI’s estimate that five U.S. hyperscalers owned 71% of the global AI compute capacity as of the fourth quarter of 2025. “This suggests that growth has indeed shifted toward AI factories, and from capability systems to capacity systems,” TOP500 organizers wrote . Benchmark and energy-efficiency leaders hold their spots LineShine also held the High Performance Conjugate Gradient (HPCG) lead at 22 petaflops, followed by El Capitan at 17.41 petaflops and Fugaku at 16 petaflops. El Capitan led the HPL-MxP mixed-precision benchmark at 16.7 exaflops, ahead of Aurora at 11.6 and Frontier at 11.4. The entire Green500 top 10 held its June positions as well. France’s KAIROS remained No. 1 at 73.28 gigaflops per watt, followed by France’s ROMEO-2025 at 70.91 and Germany’s Levante GPU extension at 69.43. The November 2026 TOP500 top 10 Six of the top 10 use HPE Slingshot-11. JUPITER Booster and Eagle use Nvidia InfiniBand, while LineShine and Fugaku use their own interconnects. 1. LineShine (China) Location: National Supercomputing Centre in Shenzhen (NSCS), China HPL score: 2.198 exaflops System model: LingKun platform Compute architecture: LX2 304-core 1.55GHz processors Total core count: 13,789,440 Interconnect: LingQi 2. El Capitan (U.S.) Location: Lawrence Livermore National Laboratory, California HPL score: 1.809 exaflops System model: HPE Cray EX255a Compute architecture: AMD 4th Gen EPYC 24-core 1.8GHz processors and AMD Instinct MI300A accelerators Total core count: 11,340,000 Interconnect: Cray Slingshot-11 width="1024" height="576" sizes="auto, (max-width: 1024px) 100vw, 1024px"> Lawrence Livermore National Laboratory/Garry McLeod 3. Frontier (U.S.) Location: Oak Ridge National Laboratory, Tennessee HPL score: 1.353 exaflops System model: HPE Cray EX235a Compute architecture: AMD Optimized 3rd Gen EPYC 64-core 2GHz processors and AMD Instinct MI250X accelerators Total core count: 9,066,176 Interconnect: Cray Slingshot-11 4. Aurora (U.S.) Location: Argonne Leadership Computing Facility, Illinois HPL score: 1.012 exaflops System model: HPE Cray EX – Intel Exascale Compute Blade Compute architecture: Intel Xeon CPU Max Series processors and Intel Data Center GPU Max Series accelerators Total core count: 9,264,128 Interconnect: Cray Slingshot-11 5. JUPITER Booster (Germany) Location: Jülich Supercomputing Centre, Germany HPL score: 1.000 exaflops System model: BullSequana XH3000 Compute architecture: Nvidia GH200 Grace Hopper Superchips with 72-core Grace CPUs at 3 GHz Total core count: 4,801,344 Interconnect: Quad-Rail Nvidia InfiniBand NDR200 6. HPC7 (Italy) Location: Eni S.p.A., Italy HPL score: 571.5 petaflops System model: HPE Cray EX255a Compute architecture: AMD 4th Gen EPYC 24-core 1.8GHz processors and AMD Instinct MI300A accelerators Total core count: 3,461,472 Interconnect: Cray Slingshot-11 7. Eagle (U.S.) Location: Microsoft Azure cloud infrastructure HPL score: 561.2 petaflops System model: Microsoft NDv5 Compute architecture: Intel Xeon Platinum 8480C 48-core 2GHz processors and Nvidia H100 accelerators Total core count: 2,073,600 Interconnect: Nvidia InfiniBand NDR 8. HPC6 (Italy) Location: Eni S.p.A., Ferrera Erbognone, Italy HPL score: 477.9 petaflops System model: HPE Cray EX235a Compute architecture: AMD Optimized 3rd Gen EPYC 64-core 2GHz processors and AMD Instinct MI250X accelerators Total core count: 3,143,520 Interconnect: Cray Slingshot-11 9. Fugaku (Japan) Location: RIKEN Center for Computational Science, Kobe, Japan HPL score: 442.01 petaflops System model: Fugaku Compute architecture: Fujitsu A64FX 48-core 2.2GHz processors Total core count: 7,630,848 Interconnect: Tofu Interconnect D 10. Alps (Switzerland) Location: Swiss National Supercomputing Centre (CSCS), Switzerland HPL score: 434.9 petaflops System model: HPE Cray EX254n Compute architecture: Nvidia Grace 72C 3.1GHz, Nvidia GH200 Superchip Total core count: 2,121,600 Interconnect: Cray Slingshot-11

Источник: Network World