← Back to feed

China’s LineShine supercomputer in Shenzhen took the #1 position on the 67th TOP500 list, revealed at ISC 2026 in Hamburg this week. It is the first Chinese system to top the list in the benchmark’s history, and the first Chinese TOP500 submission in nine years. It got there without a single GPU.

The Numbers

LineShine runs entirely on the LX2, a domestically designed Armv9-compliant CPU with SVE2 and SME support. Each chip contains 304 active cores at 1.55 GHz, delivering 60.3 TFLOP/s of FP64 compute at 690 watts per package. Each LX2 is paired with 32 GB of on-package high-bandwidth memory (likely an indigenous Chinese development rather than standard HBM) and 256 GB of DDR5 as a spillover tier, with 4 TB/s of bandwidth.

The full system: 22,000+ nodes, 13 million CPU cores, 90 compute cabinets, 800 Gbps networking per node. Sustained FP64 throughput: 2.198 Exaflops (Rmax) from a theoretical peak of 2.735 Exaflops (Rpeak), at 42.22 megawatts of power. Efficiency lands at 52.07 gigaflops per watt — lower than GPU-based systems, where leaders hit 73+ GFLOPs/W, but striking for a CPU-only architecture at this scale.

El Capitan, the previous holder, delivered around 1.742 Exaflops Rmax. LineShine beats it by roughly 26%.

HPCG: Not Just a LINPACK Special

Prior Chinese supercomputers that appeared on TOP500 were sometimes characterised as tuned specifically for the High Performance LINPACK benchmark. LineShine is not. It also took the #1 position on HPCG — the High Performance Conjugate Gradient benchmark, which is designed to be more representative of real scientific workloads. LineShine posted 22.004 Petaflops/s on HPCG, ahead of El Capitan’s 17.406. If the system runs real science as well as the HPCG result implies, the capability gap is genuine.

Export Controls Did Not Stop This

The US has imposed successive rounds of export controls on advanced semiconductors to China since 2022. LineShine is a direct data point on those controls’ limits. The LX2 chip is domestic. The networking is domestic. There is no NVIDIA, no AMD, no TSMC node that required a US export licence. The system Wired describes as “defying US restrictions” is accurate in spirit: China built a world-class machine without accessing controlled technology.

What the controls did affect, however, is the GPU-based AI training side. LineShine is an HPC system — suited for scientific simulation, weather modelling, materials science — not a generative AI training cluster. The two use cases have different chip requirements. The LineShine result does not mean export controls have no effect on Chinese AI labs; it means China can build peak HPC systems independently of US hardware.

The Hidden Exascale Fleet

China is believed to have at least two other exascale systems: Sunway Oceanlight and CNIS. Neither has appeared on a TOP500 list. The timing of LineShine’s submission — China’s first in nine years — raises a question: why now? One interpretation is that LineShine is a signal, not a complete accounting of Chinese exascale capability.

Implications for US DOE

Analysts and observers at ISC 2026 were already noting that LineShine may accelerate funding for US Department of Energy supercomputing programs. The US’s next frontier systems — including planned exascale successors to El Capitan and Frontier — are in development but not yet commissioned. If Congress interprets LineShine as a strategic gap, budget pressure on DOE exascale could shift quickly.

Key Numbers

System	Country	Rmax (Exaflops)	Architecture
LineShine	China	2.198	CPU only (LX2)
El Capitan	USA	~1.742	AMD Instinct MI300A
Frontier	USA	1.194	AMD Instinct MI250X

HPCG: LineShine 22.004 Petaflops/s vs El Capitan 17.406.

The last time a Chinese system led the TOP500 was never — LineShine is the first. The list has been US- or Japan-led for virtually its entire modern era. That changed this week, in Hamburg, with a CPU.