Core Insights - The core point of the article is the launch of the national supercomputing internet core node in Zhengzhou, which utilizes the scaleX WanKa super cluster provided by Inspur (中科曙光) to offer over 30,000 AI acceleration cards, marking a significant advancement in domestic computing infrastructure [1][4]. Group 1: Infrastructure Development - The launch of the core node validates the engineering capabilities of domestic computing infrastructure, transitioning from early single-point breakthroughs to large-scale deployment [1]. - The scaleX WanKa super cluster is a product of the "AI Computing Open Architecture" strategy proposed by Inspur, emphasizing collaboration and decoupling among various industry players [4][5]. - The system incorporates advanced technologies such as ultra-high-density blades and immersion phase change cooling, increasing computing density by 20 times and reducing Power Usage Effectiveness (PUE) to 1.04 [4]. Group 2: Market Needs and Compatibility - The market demands efficient, secure, and stable intelligent computing infrastructure, which is essential for the advancement of the AI industry [3]. - The system's open architecture allows for heterogeneous deployment of different brands of domestic acceleration cards and compatibility with mainstream computing ecosystems like CUDA, facilitating broader developer access [5][6]. - This compatibility reduces the barriers for developers and provides users with more choices, aligning with the goal of integrating computing power into industrial workflows [6]. Group 3: Performance and Applications - The WanKa super cluster supports the training and fault recovery of trillion-parameter models, enhancing inference efficiency for major internet companies [7]. - It has contributed to significant advancements in AI for Science, helping domestic material research models achieve top rankings and improving protein research efficiency by 3 to 6 orders of magnitude [7][8]. - The integration of computing power, data, and application scenarios is crucial for transforming technology into a driving force for economic development [8]. Group 4: Future Outlook - Looking ahead to 2026, the intelligent wave is expected to drive the computing industry into a new development cycle, with Inspur committed to an open technological route [8][9]. - The company aims to focus on innovation across the entire value chain, addressing the challenges of heterogeneous computing power and enhancing resource utilization efficiency [9].

Core Insights - The launch of the national supercomputing internet core node in Zhengzhou marks a significant advancement in domestic computing infrastructure, capable of providing over 30,000 AI acceleration cards [2] - The AI industry has seen a dramatic increase in computational demands, with model parameters growing from millions to trillions, leading to exponential growth in computational requirements [3] - The domestic computing industry has faced structural contradictions, with high demand for stable computing power on one side and severe fragmentation on the supply side [4] Industry Challenges - Different manufacturers have developed their own hardware designs, software stacks, and interconnection protocols, leading to a closed technology route that complicates resource scheduling and increases user migration costs [5] - The market requires efficient, secure, and stable intelligent computing infrastructure [6] Strategic Direction - In 2025, the company proposed an "AI Computing Open Architecture" strategy focused on division of labor and collaboration [7] - The scaleX supercluster launched in Zhengzhou is a product of this strategy, overcoming various technical challenges such as hardware-software optimization and high-density integration [8] Technological Innovations - The scaleX supercluster utilizes ultra-high-density blade technology and immersion phase change cooling, increasing computing density by 20 times and reducing Power Usage Effectiveness (PUE) to 1.04 [8] - The system supports heterogeneous deployment of different brands of domestic acceleration cards and is compatible with mainstream computing ecosystems like CUDA, achieving optimization for over 400 major models [9] Application and Impact - The supercluster can support the training and fault tolerance of trillion-parameter models, enhancing the efficiency of core intelligent business operations for leading internet companies [10][11] - It has significantly improved research efficiency in fields like materials science and protein research, demonstrating the potential of integrated computing, data, and application scenarios to drive economic development [11][12] Future Outlook - The company aims to continue promoting an open, efficient, and secure direction for domestic intelligent computing infrastructure, focusing on innovation across the entire supply chain [13][15] - The emphasis will be on solving the adaptation challenges of heterogeneous computing power and enhancing resource utilization efficiency [13]

Core Insights - Zhengzhou has become a strategic hub for AI computing infrastructure in China with the launch of the first domestic AI computing pool featuring 30,000 AI acceleration cards [1][12] - The deployment of the scaleX supercluster systems marks a significant milestone in the evolution of national supercomputing infrastructure, facilitating AI innovation and ecosystem empowerment [1][12] Infrastructure Development - The transition to the AI era necessitates a profound transformation in infrastructure, driven by explosive growth in demand for AI models and applications [3][14] - The national supercomputing internet aims to create a leading national-level computing facility and service platform, akin to an e-commerce marketplace for computing resources [5][16] Technical Specifications - The scaleX supercluster can provide over 30,000 AI cards for various applications, including AI model training and high-throughput inference [6][16] - The system features advanced technologies such as high-density blade servers and immersion cooling, achieving a power usage effectiveness (PUE) of 1.04 and a 20-fold increase in computing density [7][16] Innovation and Ecosystem - The scaleX supercluster represents a comprehensive innovation in system architecture and engineering, having been operationally deployed just two months after its initial reveal [8][17] - The open architecture of the scaleX supercluster supports compatibility with mainstream software ecosystems and mixed deployments of various AI acceleration cards, fostering collaboration and innovation across the AI industry [10][19] Future Implications - The open architecture approach is expected to lower barriers for users, stimulate innovation, and enable widespread application of AI across various sectors [10][19] - The deployment of the scaleX supercluster is a clear expression of China's path in developing AI infrastructure, balancing self-innovation with openness and inclusivity [12][20]

Core Viewpoint - The competitive rules of China's AI industry are being rewritten, shifting the focus from hardware performance to ecological collaboration efficiency as a new measure of competitiveness in the AI computing industry [1][4]. Group 1: Industry Trends - The HAIC2025 event showcased over 2,500 upstream and downstream enterprises, indicating a trend towards breaking down "technology walls" and "ecological walls" to achieve a "group leap" in the AI computing industry [1][3]. - The concept of "open architecture" is becoming essential for overcoming industry bottlenecks, as it promotes shared key technologies and lowers the barriers for research and application [4][6]. Group 2: Open Architecture and Collaboration - The AI computing open architecture aims to transform the industry from "physical stacking" to "chemical fusion," addressing issues such as high-end computing power shortages and high innovation thresholds [6][10]. - The launch of the scaleX super cluster, capable of deploying 10,240 AI accelerator cards and exceeding 5 EFlops in total computing power, demonstrates the technical feasibility and advantages of the open architecture [4][6]. Group 3: Partner Practices and Innovations - Companies like Qingdao Thunder Technology and Unisplendour have successfully utilized the open architecture to lower R&D costs and enhance product compatibility, leading to innovations in various industry scenarios [7][8]. - The collaborative model of "multi-vendor cooperation and shared foundation support" has proven effective in driving the domestic AI ecosystem into a "group leap" phase, with significant advancements in sectors like gaming and healthcare [7][9]. Group 4: Future Outlook - The establishment of the "AI Computing Open Architecture Joint Laboratory" aims to invest 1 billion yuan over three years, involving over 150 member units and 1,000 R&D personnel to enhance domestic AI capabilities [12]. - The ongoing reinforcement of the open architecture's "linking" role is expected to lead to a higher quality "group leap" in the domestic AI ecosystem, providing solid support for the implementation of the "Artificial Intelligence +" strategy [12].

Core Insights - The development of AI large models requires significant resources, including a large number of technical experts and substantial financial investment, with a critical need for powerful computing capabilities [1] - The demand for computing power is expected to grow exponentially across various industries, with IDC predicting that China's intelligent computing power demand will reach 2781 EFLOPS by 2028, reflecting an annual growth rate of 46.2% [1] - Traditional computing clusters face bottlenecks when scaling beyond thousands of cards, necessitating innovative solutions like the "ten-thousand card super cluster" [2] Group 1: ScaleX Ten-Thousand Card Super Cluster - The ScaleX ten-thousand card super cluster system was unveiled by Sugon at the HAIC2025 conference, designed to meet the extreme demands of AI infrastructure [3] - This system features 16 super nodes connected by a proprietary high-speed network, capable of supporting 10,240 AI accelerator cards, marking a significant advancement in domestic large-scale computing cluster technology [5] - The ScaleX system achieves a total computing power exceeding 5 EFLOPS, with a power usage effectiveness (PUE) value as low as 1.04, enhancing computing density by 20 times [5][9] Group 2: Technical Advantages - The ScaleX system utilizes a self-developed RDMA high-speed network, achieving 400 Gb/s bandwidth and under 1 microsecond communication latency, significantly improving communication performance [9] - The system incorporates deep optimization for storage, computing, and transmission, enhancing resource utilization by 55% during large model training [9] - It features a digital twin for intelligent scheduling and management, ensuring 99.99% availability and supporting the management of tens of thousands of nodes [9] Group 3: Open Architecture and Ecosystem Development - The ScaleX super cluster supports multiple brands of accelerator cards and mainstream computing ecosystems, promoting an open architecture for AI computing [10] - This initiative aims to lower the barriers for AI companies to develop intelligent computing clusters and foster a collaborative industrial ecosystem [10][12] - The open model allows users greater choice and compatibility with mainstream AI development frameworks, facilitating broader participation in the ecosystem [12][13]

2025年12月18日，在昆山举行的光合组织2025人工智能创新大会（HAIC2025）上，中科曙光发布并展出了全球领先的大规模智能计算系统——scaleX万卡 超集群，这也是国产万卡级AI集群系统首次以真机形式亮相。 作为 "AI计算开放架构"最新重磅成果，scaleX万卡超集群可支持多品牌加速卡以及主流计算生态，并实现400+主流大模型、世界模型等适配优化。在实际应 用中，该超集群可覆盖大模型训练、金融风控、地质能源勘探及科学智能等多元场景。 "AI计算开放架构" 由中科曙光协同20多家AI产业链企业共同推出，以共享若干关键共性技术能力，依托系统工程思维推进智算集群创新。通过scaleX万卡超 集群，AI企业可降低智算集群研发门槛，并从技术"单点突围"走向产业"生态共进"，将开放理念转化为可落地普惠算力。 优势1：全球首创单机柜级640卡超节点。scaleX万卡超集群由16个曙光scaleX640超节点通过scaleFabric高速网络互连而成，可实现10240块AI加速卡部署， 总算力规模超5EFlops。作为世界首个单机柜级640卡超节点，scaleX640采用超高密度刀片、浸没相变液冷等技术，将单机 ...

经济观察网12月18日，光合组织2025人工智能创新大会(HAIC2025)在江苏昆山举行。作为国内首个聚 焦"AI计算开放架构"的行业大会，本次大会汇聚了来自芯片、服务器、存储、超节点、AIPC以及大模 型、智能体和行业应用等环节的产业链上下游机构与企业，覆盖从底层算力到系统平台再到应用落地的 完整链条。与会多方围绕开放架构下的算力协同、软硬件适配和生态共建展开交流，集中展示国产智能 计算在体系化推进中的阶段性成果。 在人工智能加速走向规模化应用的背景下，HAIC2025以"产业大集"的形式搭建跨领域协同平台，将推 动国产计算生态从单点突破向系统化、开放化演进，为智能计算产业的持续发展提供现实路径。 ...

聚焦"AI计算开放架构" 光合组织2025人工智能创新大会即将举办 编辑：王永乐 广告等商务合作，请点击这里 光合组织秘书长任京暘向媒体介绍光合组织2025人工智能创新大会相关情况。中新网记者 孙自法 摄 光合组织秘书长任京暘介绍说，光合组织2025人工智能创新大会将首次大规模集中展示国产人工智能 (AI)加速计算全栈技术，发布AI计算开放架构联合实验室行动计划，基于该架构打造的全球领先大规模 智算超集群系统也将亮相。该系统采用中国自主研发的高速互联网络技术，在算力密度与系统性能上实 现双重突破，验证了开放技术路线的可行性与竞争力。 光合组织2025人工智能创新大会将设立六大分论坛，邀请百余位专家分享前沿案例，促进产业全链协同 创新，为中国AI计算的可持续发展提供开放平台与实践样板。 据悉，光合组织自成立以来，始终以构建自主可控、开放协同的国产计算生态为使命，旨在汇聚产业全 链、共破发展瓶颈，目前已有超6000家合作伙伴，建成28个生态适配中心。(完) 来源：中国新闻网 中新网北京12月5日电 (记者 孙自法)中国国内首个以"AI计算开放架构"为核心的行业盛会——光合组织 2025人工智能创新大会(HAIC2 ...

Core Insights - The World Computing Conference held in Changsha showcased the ScaleX640 super node by Inspur, which is the world's first single-rack 640-card super node, achieving a 20-fold increase in computing density and supporting million-card-level cluster expansion, thus injecting strong domestic computing power into the development of new productivity [1][3]. Group 1: Product Features - The ScaleX640 super node consists of a single immersion liquid cooling heat exchange module and two immersion liquid cooling dedicated computer cabinets, representing a significant breakthrough as it surpasses traditional solutions that would require multiple cabinets to achieve similar computing power [3]. - The device utilizes immersion phase change liquid cooling technology, allowing it to maintain a stable operating temperature below 35 degrees Celsius, which is lower than typical gaming PCs, and achieves a Power Usage Effectiveness (PUE) of 1.04, meaning 96% of electricity is used for computing, resulting in over 30% energy savings compared to traditional air cooling solutions [4]. Group 2: Industry Impact - The ScaleX640 super node adheres to the "AI computing open architecture" concept, initiated by Inspur in collaboration with over 20 industry partners, aiming to build an open and collaborative innovation system to overcome challenges in domestic AI development, such as insufficient software-hardware synergy and fragmented ecosystems [5]. - The super node supports multi-brand acceleration cards and is compatible with mainstream computing ecosystems, enabling users to flexibly select domestic computing power for rapid deployment of trillion-parameter model training and high-throughput inference, effectively addressing the practical pain points of application adaptation and software-hardware disconnection in domestic intelligent computing [5].

Core Insights - The 2025 National College Student Computer System Capability Competition - Intelligent Computing Innovation Design Competition (Pilot Cup) successfully concluded in Hefei, emphasizing the importance of AI computing open architecture in the integration of industry, academia, and research [1][3] Group 1: Competition Overview - The Pilot Cup is the only intelligent computing track in the national competition, featuring a high prize pool and employment referral opportunities for winners [3] - This year, the competition introduced a "teaching and training competition" model, enhancing the relevance of competition topics, coverage of scenarios, and integration of education and industry [3] - The competition attracted nearly 10,000 students from over 1,200 universities, with 58 teams awarded [3] Group 2: Industry Impact - The competition addressed engineering challenges in AI implementation with three key topics: "MoE language model end-to-end efficiency optimization," "ONNX Runtime operator performance optimization," and "GMRES algorithm optimization" [3] - The Senior Vice President of Sugon, Li Bin, highlighted the launch of China's first AI computing open architecture, indicating a new phase for the Chinese AI industry and a growing demand for interdisciplinary AI talent [3] Group 3: Educational Initiatives - Sugon's computing platform is user-friendly for students, comparable to CUDA, and provides extensive free learning resources [4] - The Chief Scientist of Intelligent Computing at Sugon noted an explosive growth in AI talent demand, particularly in the era of large models, and emphasized the competition's role in expanding the audience for AI talent [4] - Sugon has consistently focused on talent education, launching various initiatives such as the Pilot Cup, developer communities, and joint laboratories to explore new models for AI talent cultivation [4]