Breaking

China's LineShine Tops TOP500, Overtaking U.S. El Capitan

June 24, 2026 at 09:41 EDT

Infra & Chips

The latest TOP500 supercomputer ranking, released on June 23, 2026 during ISC 2026, places China's new system "LineShine" at No. 1 on its debut, overtaking the previously top-ranked U.S. system El Capitan. It is the first time a Chinese machine has reached the top since Sunway TaihuLight in 2017. According to the official announcement, LineShine recorded 2.198 Exaflop/s on the standard HPL benchmark, about 20% above El Capitan's 1.809 Exaflop/s.

June 2026 · TOP500 · ISC Hamburg

China's LineShine Takes No. 1 — On CPUs Alone

A Shenzhen supercomputer built entirely on domestic Armv9 CPUs — no GPUs, no AI accelerators — sustained more than 2 Exaflop/s, dethroning the U.S. El Capitan and giving China its first TOP500 crown since 2017.

2.198 EFlop/s

HPL performance — first to sustain 2+ Exaflops without GPUs

~13.79M

Cores across 20,480 nodes of the 304-core LX2 chip

~80%

HPL efficiency vs a 2.736 EFlop/s theoretical peak

HPL: LineShine vs El Capitan

Double-precision (FP64) Linpack, in Exaflop/s

2.198×10¹⁸

LineShine

Shenzhen · CPU-only

1.809×10¹⁸

El Capitan

LLNL · EPYC + MI300A

Head to head

Metric

LineShine

El Capitan

HPL (FP64)

2.198 EFlop/s

1.809 EFlop/s

Compute

CPU-only (LX2)

EPYC + MI300A

Cores

~13.79M

~11.34M

Power efficiency

52.07 GFlops/W

60.94 GFlops/W

Where it leads

1st on HPL — first 2+ EFlop/s without any GPU
1st on HPCG at 22.00 HPCG-PFlop/s
First China top spot since TaihuLight (2017), all-domestic chips

The caveats

Only 4th on AI-like HPL-MxP (7.92 EFlop/s)
Less power-efficient than El Capitan
Ranking covers only publicly submitted systems

The takeaway

A milestone for CPU-only HPC and a statement of self-sufficiency under export controls — but a machine not tuned for AI, with classified real-world use still undisclosed.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.