TOP500超级计算机榜单：中国重回榜首，欧洲保持强劲地位
最新一期TOP500超级计算机排名榜首易主，中国的LineShine超越美国的El Capitan。该榜单反映高性能计算领域竞争格局，涉及芯片与系统供应链。
原文深度解读
本文报道了第67期TOP500超级计算机排名，中国LineShine以2.198 exaflops的HPL性能首次登顶，成为首个仅用CPU突破2 exaflops的系统，也是自2017年以来中国系统首次夺冠。美国El Capitan、Frontier和Aurora分列第二至第四，欧洲JUPITER Booster位列第五。文章还提及日本、韩国、台湾的排名，并强调全球超算竞赛主要在中美之间展开，涉及芯片、互连和系统供应链。
- LineShine安装于深圳国家超级计算中心，基于国产灵鲲架构，拥有超过1370万个处理核心，采用灵启互连，HPL性能为2.198 exaflops。
- El Capitan位于加州劳伦斯利弗莫尔国家实验室，采用HPE Cray EX255a架构，配备第四代AMD EPYC处理器和AMD Instinct MI300A加速器，HPL性能1.809 exaflops，能效60.94 gigaflops/watt。
- Frontier位于田纳西州橡树岭国家实验室，采用HPE Cray EX235a架构，使用第三代AMD EPYC处理器，总核心数9,066,176，HPL性能1.353 exaflops。
- Aurora位于伊利诺伊州阿贡领导计算设施，由Intel基于HPE Cray EX架构构建，结合Intel Xeon CPU Max系列和Intel Data Center GPU Max系列，HPL性能1.012 exaflops。
- JUPITER Booster位于德国于利希超算中心，是欧洲首台百亿亿次超算，采用Eviden液冷BullSequana XH3000平台，搭载NVIDIA Grace Hopper Superchips，HPL性能1.000 exaflop。
- 日本系统排名第九，韩国Nipa排名第20，台湾Nano排名第33。
供应链影响
- 中国LineShine采用完全国产技术，可能减少对进口芯片和互连的依赖，影响相关供应链的流向。
- 美国系统依赖AMD、Intel和Cray等供应商，其持续领先可能取决于这些厂商的芯片供应和先进封装能力。
- 欧洲JUPITER Booster使用NVIDIA Grace Hopper，其性能可能受限于NVIDIA的供货和出口管制政策。
- 日本、韩国和台湾的系统排名较低，可能影响其半导体和电子产业对高性能计算资源的获取，进而影响研发竞争力。
- 全球超算竞赛可能加剧各国对先进芯片、互连和冷却技术的需求，供应链的稳定性和本土化程度可能成为关键因素。
系统解读，仅作为判断线索，不构成备货、出货、涨价或投资建议。
第67期TOP500榜单根据计算性能对全球最强大的超级计算机进行排名。今年的排名引入了新的领导者：中国的LineShine，安装于深圳国家超级计算中心。TOP500榜单以exaflops（EFLOPS）衡量性能，代表每秒执行一百亿亿（10¹⁸）次浮点运算的能力。
中国的LineShine在High Performance Linpack（HPL）基准测试中达到2.198 exaflops，成为TOP500历史上第一台仅使用CPU就持续性能超过2 exaflops的超级计算机。这也是自2017年以来中国系统首次登顶。LineShine基于中国自主研发的灵鲲架构，拥有超过1370万个处理核心，并使用专有的灵启互连。
位于加利福尼亚州劳伦斯利弗莫尔国家实验室的El Capitan跌至第二位。HPE Cray EX255a系统在HPL基准测试中达到1.809 exaflops。它拥有1134万个核心，配备第四代AMD EPYC处理器（24核，1.8 GHz）、AMD Instinct MI300A加速器以及Cray Slingshot 11互连。其能效达到60.94 gigaflops/watt。
位于田纳西州橡树岭国家实验室的Frontier保持第三位，HPL性能为1.353 exaflops。该系统基于HPE Cray EX235a架构，使用第三代AMD EPYC处理器（64核，2 GHz）。它总共包含9,066,176个核心，同样依赖Cray Slingshot 11网络。
位于伊利诺伊州阿贡领导计算设施的Aurora保持第四位，HPL得分为1.012 exaflops。Aurora由Intel基于HPE Cray EX – Intel Exascale Compute Blade架构构建，结合了Intel Xeon CPU Max系列处理器和Intel Data Center GPU Max系列加速器，通过Cray Slingshot-11互连连接。
欧洲保持竞争力
欧洲凭借JUPITER Booster保持了其在全球HPC领导者中的地位，该机器排名第五，HPL性能为1.000 exaflop。该系统是欧洲首台百亿亿次超级计算机。它安装在德国于利希超算中心，运行在Eviden的液冷BullSequana XH3000平台上，由NVIDIA Grace Hopper Superchips提供动力。
尽管取得了这一成就，欧洲仍然落后于新的中国领导者LineShine（2.198 exaflops）以及美国三大领先系统：El Capitan（1.809 exaflops）、Frontier（1.353 exaflops）和Aurora（1.012 exaflops）。
第67期TOP500榜单证实，全球超级计算竞赛已成为美国与中国之间的主要竞争。中国凭借LineShine重新夺回领先地位，这是自2017年以来中国系统首次领跑榜单。与此同时，美国仍然是高性能计算领域的主导力量，全球最快的五台超级计算机中有三台——El Capitan、Frontier和Aurora——运行在美国国家实验室。
中国以外的亚洲排名较低
其他亚洲国家也出现在榜单中，尽管排名靠后。日本凭借其先进的研究基础设施和长期积累的超级计算专业知识，继续保持强劲地位。然而，随着Fugaku统治地位的结束，日本系统现在排名第九，不再争夺第一名。
韩国排名最高的系统Nipa位列第20位，而台湾的Nano排名第33位。两国都运营着现代化的HPC中心，主要支持半导体、电子和人工智能行业。然而，就原始计算性能而言，他们的系统目前不在世界顶级梯队之列。
最新一期TOP500榜单标志着全球超级计算竞赛进入新阶段。中国、美国和欧洲首次同时运行百亿亿次系统，而世界上最快的机器完全基于中国自主研发的技术。对于欧洲而言，维持EuroHPC计划对于保持竞争力至关重要，而对于波兰来说，关键挑战仍然是建设能够参与百亿亿次计算最高级别竞争的计算基础设施。
从制造和设计到供应链和新兴技术，电子行业永不停歇。在即将举行的Evertiq Expo活动中，与推动行业发展的专业人士会面。下一站是9月17日的Evertiq Expo哥德堡，行业专业人士将齐聚一堂，交流知识、发现新解决方案并建立宝贵的商业联系。
要了解最新动态，请订阅新闻通讯。
The 67th edition of the TOP500 list ranks the world's most powerful supercomputers based on their computing performance. This year's ranking introduces a new leader: China's LineShine, installed at the National Supercomputing Center in Shenzhen. The TOP500 list measures performance in exaflops (EFLOPS), representing the ability to perform one quintillion (10¹⁸) floating-point operations per second.
China's LineShine achieved 2.198 exaflops in the High Performance Linpack (HPL) benchmark, becoming the first supercomputer in TOP500 history to exceed two exaflops of sustained performance using only CPUs. It is also the first Chinese system to top the ranking since 2017. LineShine is based on China's domestically developed LingKun architecture, features more than 13.7 million processing cores, and uses the proprietary LingQi interconnect.
El Capitan, located at Lawrence Livermore National Laboratory in California, dropped to second place. The HPE Cray EX255a system delivered 1.809 exaflops in the HPL benchmark. It features 11.34 million cores, fourth-generation AMD EPYC processors with 24 cores running at 1.8 GHz, AMD Instinct MI300A accelerators, and the Cray Slingshot 11 interconnect. Its energy efficiency reaches 60.94 gigaflops per watt.
Frontier, operating at Oak Ridge National Laboratory in Tennessee, remains in third place with an HPL performance of 1.353 exaflops. The system is based on the HPE Cray EX235a architecture and uses third-generation AMD EPYC processors with 64 cores clocked at 2 GHz. It incorporates a total of 9,066,176 cores and also relies on the Cray Slingshot 11 network.
Aurora, installed at the Argonne Leadership Computing Facility in Illinois, retained fourth place with an HPL score of 1.012 exaflops. Built by Intel on the HPE Cray EX – Intel Exascale Compute Blade architecture, Aurora combines Intel Xeon CPU Max Series processors with Intel Data Center GPU Max Series accelerators connected through the Cray Slingshot-11 interconnect.
Europe remains competitive
Europe has maintained its place among the global HPC leaders thanks to JUPITER Booster, which ranks fifth with an HPL performance of 1.000 exaflop. The system is Europe's first exascale supercomputer. Installed at the Jülich Supercomputing Centre in Germany, it operates on Eviden's liquid-cooled BullSequana XH3000 platform powered by NVIDIA Grace Hopper Superchips.
Despite this achievement, Europe still trails the new Chinese leader LineShine (2.198 exaflops) as well as the three leading U.S. systems: El Capitan (1.809 exaflops), Frontier (1.353 exaflops), and Aurora (1.012 exaflops).
The 67th TOP500 edition confirms that the global race in supercomputing has become primarily a competition between the United States and China. China has regained the top position with LineShine, marking the first time since 2017 that a Chinese system leads the ranking. At the same time, the United States remains the dominant force in high-performance computing, with three of the world's five fastest supercomputers—El Capitan, Frontier, and Aurora—operating in U.S. national laboratories.
Asia beyond China ranks lower
Other Asian countries also appear in the ranking, although further down the list. Japan continues to maintain a strong position thanks to its advanced research infrastructure and long-standing expertise in supercomputing. However, following the end of Fugaku's dominance, the Japanese system now ranks ninth and is no longer competing for first place.
South Korea's highest-ranked system, Nipa, is placed 20th, while Taiwan's Nano ranks 33rd. Both countries operate modern HPC centers that primarily support the semiconductor, electronics, and artificial intelligence industries. Nevertheless, their systems are currently outside the world's top tier in terms of raw computing performance.
The latest TOP500 edition marks a new phase in the global supercomputing race. For the first time, exascale systems are operating simultaneously in China, the United States, and Europe, with the world's fastest machine built entirely on domestically developed Chinese technology. For Europe, maintaining the EuroHPC programme will be crucial to preserving competitiveness, while for Poland the key challenge remains building computing infrastructure capable of competing in the highest league of exascale computing.
From manufacturing and design to supply chains and emerging technologies, the electronics industry never stands still. Meet the people driving it forward at the upcoming Evertiq Expo events. The next stop is Evertiq Expo Gothenburg on September 17, where industry professionals gather to exchange knowledge, discover new solutions and build valuable business connections.
To stay up to date with the latest developments, sign up for the Newsletter.