China's LineShine supercomputer achieves 2.198 exaflops on HPL benchmark, claiming top spot on the 67th TOP500 list with CPU-only architecture, displacing US-designed El Capitan.

China's LineShine supercomputer has claimed the top position on the 67th-edition TOP500 list, posting 2.198 exaflops on the High Performance Linpack benchmark and displacing the AMD-powered El Capitan into second place by more than 20%. Installed at the National Supercomputing Centre in Shenzhen and built by the Shenzhen Cloud Computing Center, LineShine achieved this ranking without GPUs or accelerators of any kind, using 13,789,440 cores of domestically designed silicon. It is the first machine on the list to exceed two exaflops of double-precision performance on CPUs alone, and the first China-based system to lead the TOP500 since Sunway TaihuLight in 2016.

The achievement carries strategic significance beyond performance metrics. A sanctioned country has successfully built an exascale flagship without any Western accelerators, but more telling is China's decision to submit the result at all. For years, its fastest systems remained off the rankings entirely; this public disclosure represents a deliberate change in posture.

LineShine is built on the LingKun platform, with 20,480 compute nodes, each carrying two LX2 processors. These Armv9-based parts feature 304 cores running at 1.55 GHz, organized as eight clusters of 38 cores. Each core includes Arm's Scalable Vector Extension and Scalable Matrix Extension units covering FP64, FP32, BF16, FP16, and INT8. Each LX2 pairs 32 GB of on-package HBM rated at up to 4 TB/s with up to 256 GB of off-package DDR5—an architecture closer to Fujitsu's A64FX in Japan's Fugaku than to conventional server CPUs. Nodes connect via the proprietary LingQi interconnect, and the system runs the homegrown Kylin OS.

The LX2's designer remains officially unconfirmed, though Jon Peddie Research has attributed the chip to Huawei, with the project's pilot phase reportedly running on Huawei Kunpeng servers. The fabrication node and foundry are similarly undocumented; SMIC's 7nm-class process is the obvious domestic candidate by process of elimination, given that EUV tooling and TSMC capacity are unavailable, but the details have not been publicly verified.

While LineShine dominated Linpack, its performance across other benchmarks reveals limitations. It claimed first place on HPCG, the test rewarding memory- and communication-bound workloads closer to real scientific code, at 22.00 petaflops. However, on HPL-MxP—the mixed-precision benchmark approximating AI training math—it ranked only fourth at 7.92 exaflops, a 3.6-times uplift over its FP64 score. In contrast, El Capitan posts 16.7 exaflops on HPL-MxP, a 9.2-times increase over its standard Linpack result, with Aurora and Frontier showing similar multipliers. This divergence exposes a fundamental difference: reduced-precision throughput is where GPUs and accelerated processors separate from CPUs, and LineShine cannot compete in this domain.

Power efficiency tells a similar story. LineShine consumes 42,220 kW and returns 52.07 gigaflops per watt on its Linpack run—surpassing Intel's Aurora but trailing El Capitan's 60.94 gigaflops per watt. LineShine produces more total FP64 output than the Livermore system while burning approximately 42% more power to do so. This distinction matters because the TOP500 ranking is decided on FP64 Linpack, the one regime where a wide, HBM-fed CPU can still match accelerators. LineShine is a genuine double-precision champion, but not a world-leading AI training machine, as its fourth-place HPL-MxP result confirms.

China's absence from recent TOP500 rankings has historical roots. The country stopped submitting its fastest systems around 2021, following entity-list additions targeting Sunway's Wuxi center and Sugon. The international community has long suspected that China operated exascale hardware well before this submission: the Sunway successor OceanLight and the NUDT-built Tianhe-3 both appeared via Gordon Bell Prize science papers without ever appearing on the list. TOP500 co-founder Jack Dongarra stated for years that Chinese researchers told him they were not permitted to submit, with omissions driven by avoiding U.S. attention rather than any capability gap.

Last year's list, topped by AMD while Chinese HPC remained absent, was conspicuously one-sided, making LineShine's current submission a reversal of that pattern. Reports indicate the system was developed without public funding, reducing political exposure in disclosure, while the all-domestic design eliminates Western dependencies that Washington could target retroactively.

Addison Snell, chief executive of HPC analyst firm Intersect360 Research, told Reuters he found the disclosure itself more surprising than the performance, noting that the surprise lay in China's desire to submit the top-ranked result and claim recognition. Ultimately, submitting a number-one system built entirely on indigenous parts signals that the sanctions regime has not achieved the technological gap China prioritizes closing.

The TOP500 hierarchy remains largely unchanged despite the leadership transfer. The U.S. still dominates three of the top five: El Capitan at 1.809 exaflops, Frontier at 1.353 exaflops, and Aurora at 1.012 exaflops. Germany's JUPITER Booster remains the first and only European exascale system at 1.000 exaflops.

AMD underpins the majority of the accelerated field, powering 191 systems on the list—an 11% year-over-year increase—and representing 41% of this edition's new entries. The company holds three top-10 slots (El Capitan, Frontier, and the newly deployed HPC7 at Italian energy firm Eni) and contributes more than 40% of combined top-10 Linpack performance. On efficiency, AMD powers 56% of the top 50 Green500 systems, and its first Instinct MI355X deployments—two Cambridge Zenith systems in the UK—entered at positions 67 and 68.

LineShine's ranking does not diminish AMD's position, primarily because the two systems pursue different workloads. AMD's MI300A and MI355X parts are engineered for mixed-precision AI arithmetic, where LineShine ranks fourth, while Western laboratories optimize for those benchmarks, not FP64 leaderboard positions. El Capitan, Frontier, and Aurora all post HPL-MxP scores several times their Linpack results through hardware capabilities LineShine lacks. While the TOP500 crown moved to Shenzhen, it did so on a benchmark that Western labs have ceased chasing with their fastest machines.