[DigitalToday reporter Jinju Hong (홍진주)] China has reclaimed the top spot for the world’s highest-performance supercomputer for the first time since 2017. It has drawn attention for achieving the No. 1 performance without GPUs despite U.S. semiconductor and AI technology restrictions.

On June 23 local time, IT outlet Engadget reported that LineShine of China’s National Supercomputing Center was selected as the world’s fastest supercomputer after recording 2.198 EFlops. That topped the 1.809 EFlops posted by El Capitan at the United States’ Lawrence Livermore National Laboratory.

LineShine is a new system that had not been disclosed in the existing TOP500 rankings. TOP500 assessed it as the first 2-exaflop-class supercomputer based on sustainable double-precision performance achieved using CPUs only.

Unlike other top-ranked systems that rely heavily on GPU accelerators, LineShine was built around a 304-core processor designed in China. The total number of cores is about 13.79 million, and the processors run at 1.55 GHz. The full system is linked with a self-developed interconnect technology, while power consumption is about 42.2 megawatts and power efficiency was measured at 52.07 gigaflops per watt.

The result is also significant because it came as U.S. semiconductor restrictions on China and limits on AI chip exports continue. Jack Dongarra of TOP500, in an interview with the New York Times, described LineShine as an "impressive system" and said it "developed a system that does not rely on GPUs and surpassed us."

With the ranking reshuffle, the number of supercomputers exceeding the exascale threshold increased to 5. LineShine in China ranked first, followed by El Capitan in the United States in second, while Frontier at the United States’ Oak Ridge National Laboratory fell to third with 1.353 EFlops.

Aurora at the United States’ Argonne National Laboratory held fourth place with 1.012 EFlops. Jupiter Booster at Germany’s Juelich Supercomputing Center ranked fifth with 1 EFlops.

The industry assesses the result as showing that a GPU-centric path is not necessarily required for supercomputer development. TOP500 explained that a range of architectures, including Intel, AMD and Nvidia, were used in top-ranked systems this year and said, "There is no single dominant technology path to leadership-class computing." Companies and research institutions are currently competing in next-generation high-performance computing through different approaches including CPUs, GPUs, APUs and custom accelerators.

China has often not disclosed supercomputer designs to the outside world, citing national security and technology protection. LineShine, however, was reportedly developed in a way that was not based on public funding, allowing it to be submitted to TOP500 benchmark tests.

The development team did not disclose some key details, including the CPU maker and semiconductor process. Still, it is being assessed as having demonstrated through the achievement that China can build a world-class supercomputer without GPUs.

The industry also sees the result as meaning more than a simple performance contest, calling it a challenge to a U.S.-centered, GPU-based supercomputing system. It is also assessed as an example showing that the exascale race is becoming more diversified around custom processors and independently designed technologies.

China’s fully domestically developed “LineShine” supercomputer ranked first on the global TOP500 list with a sustained double-precision performance of 2.19EFlops (10¹⁸ floating-point operations per second), according to the ISC 2026 conference held in Hamburg, Germany, on… pic.twitter.com/TREYRiVnAz