Workflow
昇腾384超节点算力集群
icon
Search documents
深圳首例昇腾384超节点算力集群落地商用
12月1日,记者从深圳市龙岗区城投集团获悉,昇腾384超节点算力集群在深圳福田河套建成并正式投 用。据了解,该项目由华为与深圳市龙岗区城投集团共建。 在此基础上,昇腾384超节点的落地,不仅为区域注入强劲的国产算力,更以"CANN全量开源+开源鸿 蒙生态"为双翼,打造自主的"算力-操作系统-应用"技术闭环。 CANN作为AI计算的底层引擎,确保昇腾芯片算力得以充分释放;开源鸿蒙则作为跨终端协同的操作系 统基座,打通算力流动的"最后一公里"。两者协同,不仅在技术层面实现软硬一体,更在生态层面推动 芯片、硬件、框架、应用等全产业链深度协作。例如,一个基于CANN优化的AI视觉模型,可借助鸿蒙 的分布式能力,无缝流转于手机、平板、IoT设备等多元终端,实现"一次开发、多端赋能"。 龙岗区城投相关负责人表示,未来,这一闭环体系将吸引全球AI算法开发者与鸿蒙应用开发者汇聚龙 岗、聚集深圳,形成技术共生、产业共融的集群效应,加速深圳打造"人工智能先锋城市",更将推动大 湾区构建AI产业生态。 (文章来源:21世纪经济报道) 除了384颗高性能NPU外,节点内整合了192颗鲲鹏最新一代高性能CPU处理能力,实现集群训练性能 ...
高交会观察:中国创新力“大爆炸”
Jing Ji Guan Cha Wang· 2025-11-17 14:08
Core Insights - The 27th China International High-Tech Achievements Fair (CIHTAF) concluded in Shenzhen, showcasing over 5,000 companies and achieving a transaction intention amount exceeding 170 billion yuan [1] - The event highlighted the integration of technology and application, emphasizing the need for clear and quantifiable commercial outcomes from innovations [1] Commercialization Focus - China Aerospace Science and Technology Corporation (CASC) showcased the Long March 8 rocket model, which aims to address the critical issue of reducing launch costs for commercial space endeavors [2] - The rocket has a payload capacity of 7 tons to a 700 km sun-synchronous orbit, indicating advancements in launch capabilities [2] - Beijing Jun Tian Hang Yu Technology Co., Ltd. presented the "Jun Tian No. 1 04A satellite," designed for urban management applications, showcasing the urgency for practical applications in the commercial space sector [2] Low-altitude Economy - Shenzhen Daohe Tongtai Robotics Co., Ltd. introduced an "air-ground integration" solution that significantly reduces inspection time for cross-sea bridges from 15 days to 3 days [3] - The solution utilizes a smart inspection platform called "Avant Hub," enabling coordinated operations between drones and ground robots, enhancing traffic efficiency by 42% and reducing carbon emissions by 18% [3][4] - The company plans to focus on smart transportation, smart parks, and smart energy sectors for short-term revenue generation, transitioning from selling equipment to providing services and platforms [4] Specialized Drone Applications - The China Academy of Launch Vehicle Technology showcased specialized drones for emergency rescue and security detection, indicating a shift in industry demand towards precise security and emergency response solutions [5] - The "Eagle" drone is designed for physical countermeasures, while the "Ming Die" emergency rescue system is equipped for critical area operations [5] Computing Power Infrastructure - The Huawei booth featured the "Ascend 384 Super Node" computing cluster, capable of 300 PFLOPS, aimed at supporting large-scale model training [6] - The cluster's architecture can enhance efficiency by over three times for specific models, addressing the growing demand for computational power [6] Energy Efficiency Solutions - Chaoju Technology Co., Ltd. presented the FusionPoD for AI liquid-cooled cabinet, addressing the energy consumption challenges associated with high-density computing [7] - The liquid cooling technology is essential for managing the power density exceeding 20 kW in AI applications, aiming to balance computational expansion with energy costs [7] Industrial IoT Systems - Beijing Aerospace Topology Technology Co., Ltd. showcased the "Ran Hong" industrial IoT system, which utilizes an open-source operating system to improve operational efficiency in the gas industry [8] - The system addresses fragmentation and coordination issues in traditional industrial control systems, demonstrating a shift towards full-stack domestic solutions [8] Robotics and Automation - Shenzhen Cyborg Robotics Technology Co., Ltd. focused on practical applications of robotics, demonstrating core component load testing rather than flashy performances [9] - The company aims to meet specific industrial needs with robots designed for precision and efficiency, moving away from merely aesthetic designs [11] - The trend in robotics is shifting towards adopting automotive manufacturing standards, ensuring higher reliability and performance in industrial applications [12]