In recent years, the top spot on the TOP500 list of the world’s fastest supercomputers has almost always been dominated by the United States. In June 2022, Frontier of Oak Ridge National Laboratory in the United States became the first system to truly achieve Exaflop-level results at HPL. Later, it was replaced by El Capitan of Lawrence Livermore National Laboratory in November 2024. Until the November 2025 list, El Capitan still ranked first.
However, this time there have been big changes in the TOP500 list in June 2026. LineShine from the National Supercomputing Center in Shenzhen, China, directly parachuted into the list and ran an HPL score of 2.198 Exaflop/s, officially pushing the American El Capitan into second place. More importantly, this machine does not rely on NVIDIA GPUs, AMD accelerators or other AI chips, but is a CPU-only system, which means it can outrun all the supercomputers on the current HPL list with only the CPU, which is really impressive.
China’s LineShine ranks first on TOP500! Relying only on the CPU, it ran 2.198 Exaflop, which also ended the US supercomputing ranking.
According to TOP500 official information, LineShine, which currently tops the list, is located at the National Supercomputing Center in Shenzhen, China, and was built by the Shenzhen Cloud Computing Center.
In the High Performance Linpack test used to rank the TOP500, this system ran 2.198 Exaflop/s, which is approximately 80% of the theoretical peak of 2.736 Exaflop/s. TOP500 also specifically mentioned that this is the first system on the list that can achieve sustained performance of more than 2 Exaflop/s in double-precision floating point operations “using only the CPU”:
In terms of specifications, LineShine uses China’s self-developed LingKun platform, uses a 304-core LX2 processor with an operating clock of 1.55GHz, a total core count of approximately 13.79 million, and is equipped with LingQi proprietary interconnect and Kylin operating system. The entire system consumes approximately 42.2MW and has an energy efficiency of 52.07 Gigaflops/Watt.
America’s El Capitan was bumped down to second place, with an HPL score of 1.809 Exaflop/s. This machine is located at the Lawrence Livermore National Laboratory in the United States and uses the HPE Cray EX255a architecture, paired with AMD’s fourth-generation EPYC processor and AMD Instinct MI300A accelerator.
In other words, El Capitan is a CPU plus accelerator design, while LineShine uses a CPU-only architecture to rank first this time, which is very impressive.
In fact, China has not put the fastest flagship system on the TOP500 for many years.
In the past few years, there have been no Chinese systems on the TOP500 list.. On the June 2022 list, Sunway TaihuLight is still ranked 6th, Tianhe-2A is also ranked 9th:
By November 2023,HPCwire It was also mentioned that China still had 104 systems in the TOP500, which was the second largest participating country after the United States at the time. However, it had dropped significantly from the 134 systems in June of the same year, and only one new Chinese system entered the list in that period.
In other words, Chinese systems have always been in the TOP500, but what is missing is the new generation of Exascale-level flagship models that can challenge the top spot.
This HPCwire report mentioned, China actually already has Exaflop-class supercomputers, but they are not included in the TOP500. Jack Dongarra also said that the outside world knows that China has these machines and has been operating for some time, but China has not reported official benchmark results.
Although China’s LineShine ranks first in HPL, it does not mean that it is also ranked first in the world in AI training.
LineShine won the first place this time, which cannot be simplified to “China’s CPU supercomputer comprehensively defeated all AI supercomputers.”
The TOP500 ranking mainly depends on HPL. This test focuses on double-precision floating point operations and is often used to represent traditional scientific computing capabilities. LineShine is indeed the first here, and it also scored 22.00 HPCG-Petaflop/s in the HPCG test, ranking first in HPCG.
But when it comes to HPL-MxP, which is the mixed precision operation test, LineShine’s score is 7.92 Exaflop/s, ranking only fourth.
TOP500 officials pointed out that this 3.6 times improvement is relatively limited, and is also consistent with a CPU-only design without a dedicated low-precision accelerator. In comparison, El Capitan achieved 16.7 Exaflop/s in HPL-MxP, still ranking first; Aurora and Frontier also ranked ahead of LineShine.
To put it bluntly, LineShine is really strong in traditional high-precision scientific calculations, so strong that it can squeeze out El Capitan. But if the scene is replaced by low-precision or mixed-precision operations common in AI training, the advantages of GPU, APU, and dedicated accelerators will reappear.
Source: KOCPC Chinese