Workflow
高速无损网络技术
icon
Search documents
为何要实现算力互联互通?项立刚:有助于降低训练成本、盘活算力市场,打破高价芯片垄断
Mei Ri Jing Ji Xin Wen· 2025-05-30 16:01
Core Viewpoint - The Ministry of Industry and Information Technology (MIIT) has issued the "Computing Power Interconnection Action Plan," aiming to achieve standardized interconnection of public computing power across the country by 2028, forming a computing power internet with intelligent perception and real-time discovery [1][2]. Group 1: Definition and Importance of Computing Power Interconnection - Computing power interconnection aims to break the limitations of different models lacking integration, allowing various chips to execute any model, thus enhancing the potential of computing resources [2][3]. - The action plan emphasizes the need for a standardized interconnection system to improve the efficiency and service level of public computing resources, promoting high-quality development in computing power [2][3]. Group 2: Infrastructure and Technical Requirements - The action plan outlines the necessity to focus on high-performance transmission protocols and network transmission technology research, supporting efficient data entry and lossless interconnection of computing power [2][3]. - Key technologies such as computing power identification and new computing power identification gateways are to be developed to enhance diverse computing power perception capabilities [2][3]. Group 3: Market Advantages and Economic Impact - China's vast market provides diverse demands for computing power, facilitating flexible scheduling of computing resources and enhancing overall market efficiency [5][6]. - The interconnection will allow for compatibility between different manufacturers' chips, enabling the reuse of existing computing centers, thus reducing costs and increasing efficiency [4][6]. Group 4: Data Privacy and Cleaning - The action plan addresses the importance of data privacy in the interconnection process, highlighting data cleaning as a crucial step to protect personal information while allowing valuable data analysis [7]. - Data cleaning involves removing specific personal details while retaining general characteristics for analysis, ensuring privacy protection and enabling the extraction of meaningful trends [7].