LineShine
China's debut machine seized number one with 2.198 exaflops — more than 20% ahead of El Capitan. Remarkably, it hit the top spot with a CPU-only design using domestically designed LX2 processors, no GPUs, built on China's LingKun platform.
Ranked by HPL benchmark performance on the June 2026 TOP500 list, from China's 2.2-exaflop LineShine to Switzerland's Alps.
Quick answer: 1. LineShine, 2. El Capitan, 3. Frontier, 4. Aurora, 5. JUPITER Booster, 6. HPC7, 7. Eagle, 8. HPC6, 9. Fugaku, 10. Alps.
The TOP500 list, published twice a year, ranks the world's most powerful supercomputers by their sustained performance on the High Performance Linpack (HPL) benchmark. The June 2026 edition brought the biggest shake-up in years: China's LineShine debuted at number one, ending El Capitan's run at the top and bringing China back to first place for the first time since 2017. This list takes the official top 10 from that 67th edition and compares them on the single metric that matters for the ranking — Rmax, measured in petaflops — along with the hardware behind each machine.
China's debut machine seized number one with 2.198 exaflops — more than 20% ahead of El Capitan. Remarkably, it hit the top spot with a CPU-only design using domestically designed LX2 processors, no GPUs, built on China's LingKun platform.
The U.S. Department of Energy's exascale system at Lawrence Livermore re-measured at 1.809 exaflops in the June 2026 list — a 4% improvement over its debut score. Built for nuclear stockpile stewardship, it remains the most powerful system built for a single national lab.
The first verified exascale supercomputer (2022) still holds third place with 1.353 exaflops. Hosted at Oak Ridge, Frontier's AMD-powered HPE Cray EX system drives research in physics, renewable energy, and precision engineering.
Argonne's Intel-based exascale system ranks fourth at 1.012 exaflops — the first exascale machine built on Intel CPU Max and GPU Max hardware.
Europe's first exascale-class system held the 1.000-exaflop mark from its June 2025 debut. The Jülich machine's NVIDIA GH200 superchips make it the most powerful system on the continent by a wide margin.
The June 2026 list's highest new entry at 571.5 petaflops gives Italian energy giant Eni two machines in the top 10. It runs the same HPE Cray EX255a + AMD MI300A configuration as El Capitan, scaled down for industrial seismic research.
Microsoft's Azure cloud system at 561.2 petaflops is the highest-ranked hyperscale cloud machine on the list, built on NVIDIA H100 accelerators — evidence of cloud providers' growing footprint in high-performance computing.
Eni's previous flagship at 477.9 petaflops slips to eighth but keeps the energy company as the only organization with two systems in the global top 10. It pairs AMD MI250X accelerators with an HPE Cray EX235a chassis.
Japan's Arm-based flagship, once the world champion (2020–2022), now ranks ninth at 442.0 petaflops. It remains notable as one of only two CPU-only systems in the top 10, alongside LineShine; its successor, Fugaku NEXT, is planned for around 2030.
Switzerland's HPE Cray EX system at 434.9 petaflops rounds out the top 10, built on NVIDIA Grace Hopper GH200 superchips at the Swiss National Supercomputing Centre. Finland's LUMI fell out of the top 10 in this edition.
| # | Name | chip | cores | country | interconnect | site |
|---|---|---|---|---|---|---|
| 1 | LineShine | LX2 304-core CPUs (CPU-only, no accelerators) | 13789440 | China | LingQi | National Supercomputing Centre in Shenzhen |
| 2 | El Capitan | AMD EPYC CPUs + AMD Instinct MI300A APUs | 11340000 | United States | HPE Slingshot-11 | Lawrence Livermore National Laboratory |
| 3 | Frontier | AMD EPYC CPUs + AMD Instinct MI250X GPUs | 9066176 | United States | HPE Slingshot-11 | Oak Ridge National Laboratory |
| 4 | Aurora | Intel Xeon CPU Max + Intel Data Center GPU Max | 9264128 | United States | HPE Slingshot-11 | Argonne National Laboratory |
| 5 | JUPITER Booster | NVIDIA GH200 Grace Hopper Superchips | 4801344 | Germany | Quad-Rail NVIDIA InfiniBand NDR200 | Jülich Supercomputing Centre |
| 6 | HPC7 | AMD EPYC CPUs + AMD Instinct MI300A APUs | 3461472 | Italy | HPE Slingshot-11 | Eni S.p.A. |
| 7 | Eagle | Intel Xeon Platinum 8480C + NVIDIA H100 GPUs | 2073600 | United States | NVIDIA InfiniBand NDR | Microsoft Azure cloud |
| 8 | HPC6 | AMD EPYC CPUs + AMD Instinct MI250X GPUs | 3143520 | Italy | HPE Slingshot-11 | Eni S.p.A., Ferrera Erbognone |
| 9 | Fugaku | Fujitsu A64FX 48-core CPUs (CPU-only) | 7630848 | Japan | Tofu Interconnect D | RIKEN Center for Computational Science, Kobe |
| 10 | Alps | NVIDIA Grace CPUs + GH200 Superchips | 2121600 | Switzerland | HPE Slingshot-11 | Swiss National Supercomputing Centre (CSCS) |
Items are ranked by Rmax — sustained performance on the High Performance Linpack (HPL) benchmark, in petaflops per second (PFlop/s) — exactly as published in the 67th TOP500 list (June 2026). Inclusion criteria: the 10 highest-ranked submitted systems on that published list. Systems that did not submit benchmarks (e.g., some Chinese exascale machines in prior years) are not ranked, since the TOP500 only reflects submitted results. Rmax is the measured figure; Rpeak (theoretical peak) is not used for ordering. Rank 1 is the highest value (metricDirection desc).
The TOP500 is a twice-yearly ranking (published in June and November) of the world's most powerful supercomputers, based on their performance on the High Performance Linpack (HPL) benchmark. It reflects only systems whose operators submit benchmark results — some powerful machines are never listed.
Rmax is the measured, sustained performance a system actually achieves on the HPL benchmark — the number used for the TOP500 ranking. Rpeak is the theoretical peak performance if every component ran perfectly in lockstep, which never happens in practice, so Rpeak is always higher than Rmax.
An exaflop is one quintillion (10^18) floating-point operations per second. Five systems on the June 2026 TOP500 exceed 1,000 petaflops — i.e., one exaflop — making them exascale supercomputers: LineShine, El Capitan, Frontier, Aurora, and JUPITER Booster.
No. The main TOP500 ranking uses HPL, a double-precision (FP64) benchmark, which is not how AI workloads run. Mixed-precision performance relevant to AI is tracked separately by the HPL-MxP benchmark, where El Capitan and Aurora also lead.
More like this: all Technology rankings · browse every topic