China's LineShine supercomputer took the #1 position on the June 2026 Top500 list, posting 2.74 exaflops of peak theoretical performance across 13.79 million cores and scoring just under 2.2 exaflops on the High Performance Linpack benchmark.

Installed at NSC Shenzhen, the system runs an Armv9-compatible "LingKun LX2" server CPU designed by the center in conjunction with Huawei, presumably through its HiSilicon chip division. Each LX2 die carries 304 active cores, with the chip design likely including additional cores to lift yield. A proprietary "LingQi LQLink" interconnect ties the nodes together, an InfiniBand variant or, possibly, a stripped-down Ethernet derivative.

How does LineShine compare with El Capitan?

The LX2 CPU delivers FP64 throughput through SVE2 vector units and SME matrix math blocks. LineShine's HPL result puts it 21.5 percent above El Capitan, the AMD MI300A-based system at Lawrence Livermore National Laboratory that previously held the top ranking.

China now has three known exascale-class systems with submitted or discussed results: LineShine at 2.74 EF peak, Tianhe-3 at 2.05 EF at NSC Guangzhou, and OceanLight at 1.5 EF at NSC Qingdao. The analyst covering the list believes China has additional exascale machines not disclosed publicly.

What did the June 2026 list add?

The list added 44 new machines contributing 5.3 exaflops of aggregate capacity. LineShine alone accounts for 51.6 percent of that figure. Entries need 2.66 petaflops of HPL to make the list.

Why are HPC centers still buying Hopper?

Several new clusters use Nvidia "Hopper" H100 and H200 GPUs rather than the newer "Blackwell" B200 and B300 parts. Hopper remains cheaper and offers more FP64 flops per dollar. "Hopper GPUs are cheaper, and they also have more FP64 flops and more flops per dollar than the follow-on 'Blackwell' B200 and B300 GPUs," the analyst wrote. Three new clusters do adopt Blackwell.

What does the new hardware mix look like?

• 11 new machines pair Intel Xeon CPUs with Nvidia GPUs
• 9 new machines pair AMD Epyc CPUs with Nvidia GPUs
• Combined hybrid CPU-GPU share: 15.3 percent of installed flops
• 5 new CPU-only AMD Epyc clusters: 0.8 percent of new flops
• 4 new CPU-only Intel Xeon clusters: 1.8 percent of new flops
• 2 mixed discrete-AMD CPU-plus-GPU systems: 1.7 percent of new flops
• 2 AMD MI300A-based systems (El Capitan plus HPC7): 16.3 percent of new flops

The HPC7 system at Italian oil and gas company Eni ranked #6 globally and is the largest commercial supercomputer with submitted results. The MI300A combines CPU and GPU on one package.

Are AMD and Nvidia really tied?

Across 274 accelerated systems on the list (the Top500 site text cites 277; the analyst counted 274 and triple-checked), Nvidia holds 237 machines and AMD holds 32. The capacity picture narrows considerably. Nvidia aggregates 11 exaflops of peak flops; AMD aggregates 8.18 exaflops. In combined CPU-plus-GPU concurrency, Nvidia reaches 38.9 million cores; AMD reaches 35.3 million. "This is a real race, and perhaps portends the future of AI computing as well," the analyst wrote.

Is the list still tracking real HPC?

The analyst excluded cloud-provider and telco installations from the commentary, arguing those systems distort capacity figures and do not run real HPC workloads. Installation cycles clearly track product launches: June 2024 and November 2025 were weak periods, while June 2026 and November 2024 posted the strongest additions, dominated by exascale-class machines.