大会同步发布龙岗区第二批"城市+AI"应用场景清单，涵盖21个领域共424个场景。龙岗区城投与华为 签署合作协议，共建"在地+云端"算力资源池。 龙岗区发布AI CITY标杆城区解决方案，以昇腾算力为底座，打通政务与公共服务数据，聚焦城市管 理、民生服务与产业赋能。主干道通行效率提升18%，社区医院AI辅助诊断准确率超92%。"4T数字生 活空间"以四大数字权利为核心，为市民提供T级存储、网络、公共资源与智能算力支持，构建"云-网- 边-端"一体化服务底座，实现从"城市治理"向"市民服务"转型。 昇腾超节点突破传统集群通信瓶颈，训练性能达传统节点3倍，支撑千亿级大模型训练。CANN全量开 源，推动AI框架深度适配，算子自动生成技术大幅提升开发效率。未来两年计划培育200万昇腾开发 者，夯实人才基础。 大模型训练性能提升3倍 【深圳商报讯】（记者 马小晗）近日，"昇腾超节点暨CANN生态合作大会"在深圳华为坂田基地召开。 大会以"智算新纪元，生态创未来"为主题，集中发布昇腾超节点算力方案、CANN开源生态成果、AI CITY标杆城区解决方案及"4T数字生活空间"。 ...

Group 1 - The conference highlighted the importance of "AI computing power" and introduced the AI CITY benchmark solution, which aims to connect government data, public service data, and citizen needs [1] - The "4T for Home" project envisions a future where ordinary residents can benefit from digital technology, emphasizing the practical application of digital solutions in daily life [1] - The initiative aims to make digital cities more relatable and beneficial to residents, moving away from the perception of smart cities as complex systems [1] Group 2 - The "4T for Home" project promotes the concept of "digital equality," allowing all residents, regardless of their technical knowledge, to access public resources and AI tools [2] - The construction of AI CITY and the "4T for Home" project is expected to face challenges, but the value of AI technology will ultimately be measured by its impact on people's lives [2] - The initiative is supported by significant government investment and aims to create a positive ecosystem where government guidance, technological support, and resident benefits coexist [2]

2025年9月18日，华为全联接大会2025——昇腾AI人工智能产业峰会将在上海举行，本次峰会以"与时代 共昇腾"为主题，将重磅呈现昇腾从基础软硬件到行业解决方案的最新技术成果与创新突破，同时客户 伙伴将现场分享基于昇腾技术的商业落地案例与价值洞察，共享创新实践优秀经验与成果，共绘智能世 界新蓝图。 来源：环球网 ●新价值：行业实践，共享智能化跃迁新成果 在行业实践方面，峰会将重磅发布运营商、政务、教育、金融、大模型、电力六大行业大规模专家并行 优秀实践；邮储银行嘉宾也将围绕昇腾大规模专家并行解决方案在数字金融场景的创新实践进行主题分 享，为行业同仁提供可借鉴的经验模式。 本次峰会将重点呈现三大核心亮点： ●新技术：持续创新，突破AI技术新高度 峰会将深度解读昇腾"基础硬件、基础软件及行业解决方案"的最新技术进展。从业界最大规模超节点技 术的突破，到基础软件的全面开源开放，再到预训练、后训练及推理解决方案的全链路技术创新……一 系列前沿成果将集中呈现，为技术爱好者打造一场精彩纷呈的知识盛宴。 ●新方向：超节点引领AI基础建设新趋势 大模型正以不可阻挡之势推动全球计算领域实现跨越式变革，其技术与能力的持续演进引 ...

Group 1 - The fourth Global Digital Trade Expo will take place in Hangzhou from September 25 to 29, showcasing cutting-edge technologies and immersive experiences, heavily featuring artificial intelligence [1] - The expo will include a main digital trade exhibition area and seven specialized industry exhibition areas, focusing on sectors such as e-commerce, artificial intelligence, smart transportation, digital entertainment, digital healthcare, smart spaces, and smart logistics [1] - Leading companies and institutions like Alibaba, Huawei, and Zhijiang Laboratory will present their AI innovations, demonstrating capabilities from foundational models to advanced applications [1][2] Group 2 - The AI exhibition area has doubled in size this year, covering a broader industry chain and introducing a dedicated intelligent agent section, featuring the world's first intelligent agent capable of perception, interaction, and reach [2] - Over 80 robotic products will be showcased at this year's expo, significantly increasing from the previous year, with a dedicated robot show area featuring various performances [2] - The "Artificial Intelligence +" concept will be integrated throughout the exhibition, with other specialized areas presenting the latest AI products and applications across various industries, including cross-border e-commerce, green transportation, medical diagnostics, cultural entertainment, and furniture architecture [3]

Group 1 - Huawei announced the open-source release of the Pangu model with 70 billion parameters and the Pro MoE 72B model with 720 billion parameters, enhancing the development of large model technology on domestic computing platforms [1] - The Pro MoE 72B model achieves superior performance comparable to trillion-parameter models by dynamically activating a network of experts, with only 160 billion parameters activated during operation [1] - The latest Super CLUE ranking places Huawei's large model as the top domestic model within the trillion-parameter category, indicating significant advancements in the field [1] Group 2 - Huawei's Ascend chips and CANN heterogeneous computing architecture are part of a fully autonomous and optimized closed-loop solution, marking a shift in global AI computing competition towards large-scale system efficiency and ecosystem development [2] - The Ascend super node has been commercially deployed in data centers such as China Telecom, contributing to the growth of the domestic supply chain [2] Group 3 - Huasen Tiancheng has collaborated with Huawei on Ascend chips and AICC intelligent contact center initiatives, indicating a strategic partnership in AI technology [3] - Softcom Power has launched the Softcom Ascend AI workstation to enhance local AI inference and production efficiency, reflecting the industry's push towards localized AI solutions [4]

Core Insights - The Guangdong-Hong Kong-Macao Greater Bay Area is experiencing strong growth in the computing power industry, with Shaoguan becoming the largest smart computing cluster in South China [1][2] - The total computing power scale of the eight major hubs in China's "East Data West Computing" initiative reached 215.5 EFLOPS, accounting for over 70% of the national total [1] - Shaoguan has established a capacity of 120,000 standard racks and attracted 22 smart computing center projects with a total investment exceeding 60 billion [2] Industry Development - A series of technology innovation platforms were launched at the conference, including the Shaoguan Data Investment & Huawei Joint Innovation Laboratory, which will focus on data elements, talent cultivation, and adaptation centers [3] - The AI Edge Alliance was officially established to promote the integration of AI and edge computing, aiming to inject new vitality into the smart computing industry's upgrade [3] - Shaoguan's network latency has improved significantly, with data transmission to Guangzhou taking only 1.3 ms and to Shenzhen 1.66 ms, meeting most computing power business needs in the Greater Bay Area [2] Investment and Projects - The Greater Bay Area Computing Scheduling Center has been completed, and the Guangzhou Data Exchange (Shaoguan) service base is operational [2] - China Telecom plans to invest over 2.4 billion in Shaoguan to build 38,500 standard racks, while Softcom Power Information Technology will develop a computing power scheduling platform [3]

Core Viewpoint - The article discusses the advancements in AI, particularly focusing on Huawei's innovations in the MoE (Mixture of Experts) architecture and the introduction of RL (Reinforcement Learning) post-training techniques, which aim to enhance the efficiency and performance of large language models (LLMs) in the competitive AI landscape [1][3]. Group 1: MoE Architecture and Huawei's Innovations - The MoE model, originally proposed by Canadian scholars, has evolved significantly, with Huawei introducing the MoGE architecture that addresses inefficiencies in the traditional MoE model, leading to cost reduction and improved training and deployment [1]. - Huawei's approach emphasizes the importance of creating a collaborative ecosystem to foster the growth of the Ascend ecosystem in China [1]. Group 2: RL Post-Training Techniques - RL post-training has emerged as a critical pathway to enhance LLM performance, with models like OpenAI's o1 and DeepSeek-R1 leveraging this technique to improve reasoning capabilities in complex tasks [3][5]. - The RL post-training phase currently consumes 20% of the total computational resources, projected to rise to 50%, significantly impacting model performance and costs [3]. Group 3: Challenges in RL Post-Training - The traditional On-Policy algorithms create a "computational black hole" due to the alternating execution of training and inference tasks, leading to underutilization of resources [6][7]. - The complexity of task scheduling in large-scale clusters, exacerbated by the adoption of various parallel strategies, poses significant challenges for efficient resource utilization [8]. Group 4: Innovations in Resource Utilization - Huawei's RL Fusion technology allows a single card to handle both training and inference tasks simultaneously, effectively doubling resource utilization and throughput [9][10]. - The StaleSync mechanism enables near-asynchronous execution of tasks, achieving over 90% efficiency in horizontal scaling across CloudMatrix 384 super nodes [16][20]. Group 5: Performance Metrics and Results - The combination of RL Fusion and StaleSync has led to a significant increase in efficiency, with single-node throughput improving by 78.5% and overall performance enhancement of 1.5 times [30][31]. - StaleSync's implementation in cluster scaling shows a linear throughput increase from 35k tokens/s to 127k tokens/s as the number of super nodes increases, demonstrating its effectiveness in enhancing scalability [32]. Group 6: Conclusion - The advancements in RL post-training techniques by Huawei represent a significant leap in AI efficiency, positioning the company as a key player in the next generation of AI technology [33].

Core Insights - The rapid growth of large models in AI is driving a new era of computing power demand, highlighting the limitations of traditional cluster architectures in efficiently training these models [1][2] - Traditional architectures face significant challenges, including communication bottlenecks, inefficient resource allocation, and reliability issues, which hinder the training efficiency of large models [2][3] Summary by Sections Challenges in Traditional Architectures - Communication bottlenecks have worsened exponentially, with MoE models increasing inter-node communication demands, leading to delays of over 2ms in traditional 400G networks [1][2] - Resource allocation is static and unable to adapt to dynamic changes in model structure, resulting in a 30% decrease in overall training efficiency due to uneven load distribution [1][2] - Reliability is compromised as the probability of node failure increases with scale, causing significant resource waste during lengthy recovery processes, with some companies losing over a million dollars per training interruption [2] Emergence of Ascend Supernode Architecture - The Ascend Supernode architecture represents a fundamental restructuring of computing power systems, characterized by a "three-dimensional integration" approach [3][5] - A breakthrough in hardware interconnectivity allows multiple NPUs to work as a single computer, increasing inter-node communication bandwidth by 15 times and reducing latency from 2ms to 0.2ms [3][5] - Unified global memory addressing through virtualization enables direct memory access across nodes, enhancing efficiency in parameter synchronization during model training [5][6] Innovations in Resource Management and Reliability - Intelligent resource scheduling allows for fine-grained dynamic task allocation based on the MoE model structure, improving the compute-to-communication time ratio from 1:1 to 3:1 [5][6] - The reliability of the system has been significantly improved, with average uptime increasing from hours to days, and recovery times reduced from hours to 15 minutes [5][6] Industry Impact and Future Prospects - The Ascend Supernode architecture has achieved a threefold increase in training performance compared to traditional nodes, establishing a new benchmark in AI computing [8] - The introduction of MindIE Motor enhances large-scale expert parallel capabilities, achieving four times the throughput of traditional server stacks [8] - Huawei's commitment to architecture innovation is seen as a new form of Moore's Law, positioning the company as a leader in the AI computing landscape [9]

Group 1: Power Industry - The overall electricity consumption in April is expected to maintain a growth rate between 4.5% and 5.5% year-on-year, indicating a positive trend for the power sector [5] - Several companies in the power sector have shown significant stock performance, including Huayin Electric (9.91% increase), Huaneng Energy (10.00% increase), and others [5][6] - New energy companies like Xineng Taishan and Jineng Technology also reported strong stock performance, with increases of 10.14% and 9.93% respectively [4][5] Group 2: Computing Power Industry - The launch of the world's first commercial intelligent computing super node in the Guangdong-Hong Kong-Macao Greater Bay Area is expected to boost the computing power sector [7] - Companies such as Zhongdian Xilong and Tianyu Shuke have seen stock increases of 10.00% and 10.07% respectively, driven by developments in computing power [8] Group 3: Robotics Industry - UBTECH has signed a large procurement contract for industrial humanoid robots, which is likely to enhance the robotics sector [9] - Stocks like Dongbei Group have shown strong performance with a 10.00% increase, reflecting positive market sentiment in the robotics industry [9] Group 4: Chemical Industry - Recent fluctuations in chemical product prices have drawn market attention, impacting related companies positively [10] - Companies such as United Chemical and Ding Jixin have reported stock increases of 11.23% and 9.98% respectively, benefiting from the chemical market dynamics [13] Group 5: Advanced Materials - The rise of humanoid robots is expected to benefit upstream core chemical new materials, including PEEK and other high-end engineering plastics [14] - Stocks like Kent Shares and New Trend Materials have seen increases of 14.33% and 13.82% respectively, driven by the demand for advanced materials [15]

Group 1 - The core viewpoint of the article highlights a rebound in computing power concept stocks, with significant gains in companies like Dawi Technology and Hongbo Shares reaching their daily limit [1] - The article notes that several other companies, including Hongjing Technology, Hengrun Shares, Hang Steel Shares, Kehua Data, Aofei Data, and Yunsai Zhili, saw increases of over 5% [1] - A key event mentioned is the launch of the world's first commercial intelligent computing Ascend super node, which went live in the Guangdong-Hong Kong-Macao Greater Bay Area (Shaoguan) computing power cluster on April 26 [1]