但当时一些讨论认为，华为是靠堆砌了384颗芯片，才超越了英伟达72颗芯片，而后很多厂商也跟进了 超节点的概念。一时间，算力行业掀起了建设超节点的热潮。那回到最初的问题，超节点就是单纯的堆 芯片吗？华为是通过暴力堆砌芯片赶超英伟达的吗？超节点和传统的计算集群区别在哪？ 关于这些行业热点话题，华为计算产品线营销运作部部长张爱军近日在一场媒体沙龙上对观察者网等媒 体做了详解。他首先认为，今天的算力需求远未被满足，无论是大模型进入到生产系统中，还是在toC 消费端，token消耗的数量都在指数级增长，未来中国每日token消耗量可能突破千万亿。 文 观察者网 吕栋 在一些观点看来，算力缺口大可以通过大量建设服务器集群堆卡来解决。这么说听起来有一定道理，但 真实情况远没有这么简单。根据Meta公布的论文，万卡集群训练时算力利用率仅约38%，粗暴堆卡可能 会造成62%的算力浪费，并且模型训练会3个小时中断一次。 2025年以来，超节点（SuperPod）作为新的AI算力基础设施，不断成为行业焦点。观察者网也曾深度报 道过华为的昇腾384超节点，它通过高速互联总线将384颗昇腾芯片连接起来，在超节点算力规模、网络 互联带宽 ...

2025年以来，超节点（SuperPod）作为新的AI算力基础设施，不断成为行业焦点。观察者网也曾深度报 道过华为的昇腾384超节点，它通过高速互联总线将384颗昇腾芯片连接起来，在超节点算力规模、网络 互联带宽、内存总带宽等多个指标上，赶超了国外厂商的旗舰系统。 但当时一些讨论认为，华为是靠堆砌了384颗芯片，才超越了英伟达72颗芯片，而后很多厂商也跟进了 超节点的概念。一时间，算力行业掀起了建设超节点的热潮。那回到最初的问题，超节点就是单纯的堆 芯片吗？华为是通过暴力堆砌芯片赶超英伟达的吗？超节点和传统的计算集群区别在哪？ 文 观察者网 吕栋 在一些观点看来，算力缺口大可以通过大量建设服务器集群堆卡来解决。这么说听起来有一定道理，但 真实情况远没有这么简单。根据Meta公布的论文，万卡集群训练时算力利用率仅约38%，粗暴堆卡可能 会造成62%的算力浪费，并且模型训练会3个小时中断一次。 这是因为，集群网络通信已成为当前大模型训练和推理的最大挑战。以DeepSeek这种混合专家模型 （MoE）为例，每个"专家"如果不能有效沟通，NPU就会由于没有足够数据计算而闲置，进而形成 1+1<2的结果。换句话说，如果 ...

Group 1 - The openGauss community announced that openGauss has achieved a 35.02% market share in China's offline centralized relational database market, with the commercial version accounting for nearly 30% (29.4%) [1] - The community introduced the "1+2" strategy focusing on the development of oGRAC multi-read and multi-write databases, as well as an AI-native multimodal database foundation [1] - The newly released oGRAC multi-master database architecture offers RPO=0 (zero data loss) and RTO<10 seconds for business recovery, with TPCC performance reaching 3.5 million tpmC in a dual-node deployment [1] Group 2 - The Chinese Academy of Sciences Software Research Institute has officially joined the openGauss community council to enhance original innovation in the foundational software sector [2] - openGauss has gathered over 880 industry partners and more than 8,400 developers, with global downloads exceeding 5.5 million [2] - Huawei's Kunpeng Computing Business President reiterated Huawei's commitment to the community, stating that they will share technologies such as modular capabilities and distributed memory with the database ecosystem [2]

Core Insights - The openGauss Summit 2025 showcased the innovation and ecological impact of openGauss as an enterprise-level open-source database community, emphasizing its role in the evolution of open-source database technology and AI integration [1][10]. Group 1: Community and Industry Collaboration - The openGauss community has gathered over 880 industry partners and more than 8,400 global developers, with over 5.5 million downloads of its versions [6]. - The community's "1+2" strategy focuses on long-term technological evolution and aims to develop two major breakthroughs: the oGRAC multi-read and multi-write database and an AI-native multimodal database foundation [6]. - A "Supernode Database Industry-Academia-Research Alliance" has been established to promote technological innovation and results transformation in supernode architecture [9]. Group 2: Technological Advancements - The oGRAC multi-master database architecture features global consistency mechanisms and rapid failover technology, achieving RPO=0 (zero data loss) and RTO<10 seconds (business recovery in seconds) [8]. - The DataVec capability has been enhanced to support integrated AI pipelines and in-database large models, addressing issues like "model hallucination" [8]. - The first multi-database kernel engine, DataVec, has been launched, enabling precise and rapid retrieval for intelligent platforms [8]. Group 3: Market Performance - openGauss holds a 35.02% market share in China's offline centralized relational database market, with its commercial version accounting for 29.4%, making it the leading open-source database [10]. - The technology has been successfully implemented in various sectors, including finance, government, telecommunications, healthcare, manufacturing, and industrial control [10]. - Companies like Zhihu, Donghua Medical, and others shared their successful practices in performance tuning and intelligent manufacturing, validating openGauss's reliability in complex business scenarios [10].

Core Viewpoint - The establishment of the "Supernode Database Industry-Academia-Research Alliance" aims to promote breakthroughs and innovations in database technology under the supernode architecture, accelerating the transformation and application of results [1] Group 1: Alliance Formation - The alliance consists of 13 entities, including operators, databases, the Chinese Academy of Sciences, and top universities [1] - The Chinese Academy of Sciences Software Research Institute has announced its membership in the openGauss community council [1] Group 2: Community Achievements - The openGauss community has gathered over 880 industry partners and more than 8,400 global developers in five years [1] - The community version has been downloaded over 5.5 million times globally [1] Group 3: Market Position - In the Chinese offline centralized relational database market, openGauss holds a market share of 35.02% [1] - The commercial version of openGauss accounts for 29.4% of the market, surpassing other open-source database versions [1] - openGauss has maintained its position as the leading database solution in the market [1] Group 4: Future Commitment - Huawei's Kunpeng Computing Business President Li Yi stated that Huawei will continue to invest firmly in the openGauss community [1]

Core Insights - The article discusses the establishment of the "Supernode Database Industry-Academia-Research Alliance" at the openGauss community summit, highlighting the importance of databases in the AI era and the need for innovation in domestic database technology [1][2]. Group 1: Industry Trends - The rise of supernode technology is identified as a key trend, which integrates thousands of computing cards into a powerful computing unit for large-scale tasks [1]. - The Chinese database industry is at a critical stage of high-quality development, facing challenges such as technology route collaboration and the integration of AI [2]. Group 2: Community and Collaboration - The openGauss community has grown significantly, with over 880 industry partners and more than 8,400 global developers, achieving over 5.5 million downloads of its community version [2]. - The integration of top national research institutions, such as the Chinese Academy of Sciences, into the openGauss community is expected to enhance original innovation and exploration capabilities in the foundational software field [3]. Group 3: Market Position - openGauss has achieved a market share of 35.02% in China's offline centralized relational database market, with a commercial version adoption rate of 29.4% [2].

Core Insights - The event "Kunpeng Developer Day" in Beijing focused on the theme "Create the Future, Enjoy the Extraordinary," featuring a main forum, two sub-forums, and an interactive exhibition area to foster deep technical exchanges and industry practice sharing among developers from various sectors [1] Group 1: Event Overview - The event was organized by Huawei Technologies Co., Ltd. and hosted by the Beijing Kunpeng Joint Innovation Center, bringing together hundreds of developers from universities, enterprises, and research institutions [1] - Huawei's Vice President Ma Haixu expressed gratitude to developers for their support of the Kunpeng ecosystem, emphasizing their role in driving innovation and collaboration [1] Group 2: Technical Presentations - Peking University's Fan Chun discussed the integration of cloud computing and high-performance computing in AI applications, highlighting the development of the "Hesi Scheduling System" and the "SCOW Computing Platform," which supports over 80 computing centers [3] - Huawei's Li Sicong introduced the Kunpeng Unified Acceleration Library (KUPL), which enhances performance in high-performance applications by 2.7 to 4.6 times compared to traditional hardware, and reported that the openEuler operating system is expected to exceed 16 million installations by the end of 2025 [5] - Tsinghua University's Zong Zan shared advancements in large model inference acceleration on the Kunpeng architecture, achieving significant throughput improvements [7] Group 3: Industry Applications - North控数科's Cheng Xiaopei presented a cloud platform for state-owned enterprises built on Kunpeng technology, which has developed over 140 solutions across various sectors [9] - China Everbright Bank's Shen Jie highlighted the bank's use of Kunpeng's trusted cloud technology to enhance core business system performance, achieving a 36% increase in concurrency and a 27% reduction in latency [11] - Yuxin Technology's Song Tao discussed the end-to-end solutions based on Kunpeng technology, showcasing significant efficiency improvements in data processing for a banking project [13] Group 4: Educational and Collaborative Initiatives - China National Knowledge Infrastructure's Wang Liang shared the implementation of Kunpeng nodes in their research platform, resulting in reduced latency and energy costs [16] - The event featured the launch of the "Kunpeng Partner Co-Development Excellent Solutions" initiative, marking a new phase in collaborative development with over a thousand partners [18] - The Kunpeng Innovation Competition awarded prizes to outstanding teams, encouraging developers to continue exploring Kunpeng technology [20] Group 5: Developer Engagement - The event included technical forums and hands-on training sessions to empower developers with practical skills and knowledge about Kunpeng technology [23] - An interactive exhibition area provided developers with direct access to Kunpeng tools and resources, enhancing their experience and engagement [25] - The successful hosting of the Kunpeng Developer Day is expected to inject new momentum into the Kunpeng ecosystem in Beijing and North China, promoting the deep integration of computing technology into various industries [27]

Core Insights - The event "Kunpeng Developer Sharing Day" was held in Beijing, focusing on the theme "Create the Future, Enjoy the Extraordinary," featuring a main forum, two sub-forums, and an interactive exhibition area to foster deep technical exchanges and industry practice sharing among developers from various sectors [1][30]. Group 1: Event Overview - The event was organized by Huawei Technologies Co., Ltd. and hosted by the Beijing Kunpeng Joint Innovation Center, gathering hundreds of developers from universities, enterprises, and research institutions [1][30]. - Huawei's Vice President, Ma Haixu, expressed gratitude to developers for their support of the Kunpeng ecosystem, emphasizing the importance of developers in driving innovation and collaboration [1][31]. Group 2: Technical Presentations - Peking University's Director of the Computing Center, Fan Chun, discussed the integration of cloud computing and high-performance computing in AI applications, highlighting the "Hesi Scheduling System" and the "SCOW Computing Platform" that supports over 80 computing centers [3][33]. - Huawei's high-performance computing expert, Li Sicong, introduced the unified acceleration library system (KUPL), which enhances performance in various applications, achieving up to 4.6 times improvement in traditional high-performance applications [5][35]. - Tsinghua University's researcher, Zong Zan, shared optimization practices for large model inference acceleration on the Kunpeng platform, achieving significant throughput improvements [7][37]. Group 3: Industry Applications - North Control Data's General Manager, Cheng Xiaopei, presented a cloud platform developed for state-owned enterprises, which has created over 140 solutions across various sectors, including energy and manufacturing [10][39]. - China Everbright Bank's technology manager, Shen Jie, reported a 36% increase in concurrent processing capacity and a 27% reduction in latency through the implementation of Kunpeng's trusted cloud technology [12][41]. - Yuxin Technology's solution director, Song Tao, discussed the end-to-end solutions built on Kunpeng technology, showcasing significant efficiency improvements in data processing for a major bank [14][43]. - China National Knowledge Infrastructure's Deputy Director, Wang Liang, highlighted the reduction in research computing latency by 35% and power costs by 33% through the adoption of Kunpeng nodes [17][44]. Group 4: Ecosystem Development - The event featured the launch of the "Kunpeng Partner Co-Development Excellent Solutions" initiative, marking a new phase in collaborative development with over a thousand partners [19][47]. - The Kunpeng Innovation Competition awarded prizes to outstanding teams, encouraging developers to continue exploring Kunpeng technology [21][49]. - The event included technical forums and hands-on training sessions, providing developers with resources and experiences to effectively utilize Kunpeng technology [24][52].

Group 1 - The recommendation is to learn multiple database systems, specifically Dameng, Kingbase, and PostgreSQL, as they share foundational knowledge, making the learning process complementary [2] - The future DBA role will require proficiency in multiple domestic databases, as larger enterprises typically utilize at least 3-5 different systems, necessitating a broader skill set [3] - The operational requirements for DBAs have shifted significantly from traditional Oracle DBA roles, with an increased focus on basic operations and application-related tasks, indicating a need for adaptation and skill enhancement [4] Group 2 - The learning curve for domestic databases is relatively shallow compared to previous Oracle knowledge, suggesting that current training may not be as comprehensive [2] - Political factors are influencing the adoption of various database products, leading to additional learning burdens for DBAs in the industry [3] - DBAs are encouraged to proactively expand their skill sets and prepare for the evolving landscape of database management, emphasizing the importance of adaptability in their careers [4]

Core Insights - The event "2025 Global HCIE Night" successfully gathered over 200 HCIE experts, industry clients, ecosystem partners, and Huawei industry leaders from various regions including China, Europe, Asia-Pacific, Africa, Latin America, and the Middle East [1] - Huawei announced that the number of certified engineers has surpassed one million, marking a significant milestone in the industry [4] - The event highlighted the importance of AI skills in the ICT sector, with a focus on the integration of industry knowledge and AI technology [6][8] Group 1 - The event featured discussions on industry trends in intelligent computing, networking, and data storage, emphasizing the need for high-quality service capabilities [3] - Huawei and the Ministry of Industry and Information Technology jointly released the "ICT Talent Evaluation Action Plan," aiming to enhance talent standards and evaluation systems in key areas like AI [8] - The demand for talent is shifting towards a combination of industry knowledge and AI technology, with an increasing need for professionals who can integrate systems and ensure security and sustainability [6] Group 2 - Awards were presented to 40 HCIE experts and 4 partners, recognizing their contributions to the industry, including the "HCIE Ten-Year Special Contribution Award" for experts with over a decade of certification [12][25] - The event included testimonials from professionals who shared their experiences with HCIE certification, highlighting its impact on their careers and the importance of continuous learning [25][29][31] - Huawei's training and certification department reported that the number of HCIE experts has reached 37,000 globally, with ongoing efforts to expand certification offerings and enhance career development pathways [35]