Workflow
联想深腾X8810系统
icon
Search documents
对话联想ISG黄山:异构算力融合将成为AI应用落地的关键丨ToB产业观察
Tai Mei Ti A P P· 2025-08-19 02:43
Group 1 - The core viewpoint of the article highlights the transition of generative AI from technology to application, indicating that while many application scenarios have emerged, the industry is still far from achieving enterprise-level AI applications akin to the "iPhone moment" [2] - The intelligent computing industry has seen rapid growth since the advent of ChatGPT, with China's intelligent computing scale reaching 748 EFlops by Q1 2025, accounting for 35% of the overall computing scale [3] - IDC forecasts that China's intelligent computing scale will reach 1,037.3 EFlops by 2025 and 2,781.9 EFlops by 2028, with a compound annual growth rate of 46.2% from 2023 to 2028 [3] Group 2 - The demand for computing power is primarily driven by intelligent computing, with the global AI server market expected to grow from $125.1 billion in 2024 to $158.7 billion in 2025, and potentially reaching $222.7 billion by 2028 [4] - The generative AI IaaS market in China saw a year-on-year growth of 203.6% in the first half of 2024, reaching a market size of 5.2 billion RMB, which constitutes 35.6% of the overall intelligent computing service market [4] - The evolution of large model technology is shifting from pre-training to inference and post-training, indicating a change in the industry's computing demand [4][5] Group 3 - The concept of "super-intelligent fusion" is crucial for maximizing resource efficiency and supporting complex scientific problems and industrial upgrades [6] - Supercomputing, intelligent computing, and traditional data centers differ in application scenarios, with supercomputing focusing on large-scale scientific calculations and intelligent computing on AI and machine learning [6] - The integration of supercomputing and intelligent computing is expected to enhance computing efficiency and reduce energy costs, with collaborative efforts improving overall performance [6][7] Group 4 - The super-intelligent fusion approach has shown significant benefits in various applications, such as drug development and manufacturing process upgrades, with notable efficiency improvements reported [7][8] - Challenges remain in the integration of supercomputing and intelligent computing due to differences in hardware architecture and computational paradigms [8][9] - A platform-based solution is emerging as a preferred approach to address the integration challenges, with companies like Lenovo developing heterogeneous intelligent computing platforms to enhance computing resource utilization [9][10] Group 5 - Lenovo's heterogeneous intelligent computing platform has achieved significant compatibility and performance improvements, with a reported 15-fold speed increase compared to traditional CPU clusters [10] - The platform integrates various AI models and optimizes them for different architectures, becoming a core tool for enterprise-level AI deployment [10][11] - The trend towards platformization in the industry is evident, with multiple companies launching their own solutions to enhance the usability of hardware and software [11]