在7月7日举行的信息化百人会（ChinaInfo100，信百会）2024年度研讨会上，中国工程院院士、清华大学计算机科学与技术系教授郑纬民发表了引人深思的演讲。郑院士指出，当前AI大模型的发展趋势正从单模态向多模态转变，应用领域广泛，这直接导致了算力需求的爆发性增长。然而，与国际领先企业如英伟达相比，国产AI芯片在系统生态建设方面存在明显差距。
在演讲中，郑纬民强调了AI芯片生态的重要性。他提出，生态的成熟度直接关系到芯片的普及和应用程度，即便性能只达到60%，只要生态完善，仍然会有大量的用户愿意采用。这反映了生态建设对国产AI芯片的迫切需求，强调了构建一个全面、开放、兼容的生态系统对于推动国产AI芯片发展的关键性作用。
郑院士进一步分析指出，AI大模型的多模态发展，要求芯片能够处理和优化多类型数据，如文本、图像、音频和视频等，这对芯片的计算能力、内存管理和能耗控制等提出了更高要求。而与国际巨头相比，国产AI芯片在生态建设方面仍面临挑战，包括但不限于软件开发工具、应用开发平台、系统集成能力、行业解决方案等方面的不足。
针对这一现状，郑纬民呼吁行业内外共同努力，加速构建国产AI芯片的生态系统。他强调，生态建设需要政府、企业、研究机构以及广大开发者共同参与，通过提供标准化的接口、丰富的开发工具、多样化的应用案例、以及开放的交流平台，促进技术的快速迭代和应用的广泛普及。
此次演讲不仅揭示了国产AI芯片在生态建设方面的紧迫性，也为未来的发展指明了方向。通过加强生态建设，国产AI芯片有望在激烈的市场竞争中占据一席之地，为我国人工智能产业的发展注入新的活力。
英语如下：
News Title: “Domestic AI Chip Ecosystem Shortcomings Limit Performance, Experts Call for Accelerated Ecosystem Development”
Keywords: Domestic AI Chip, Ecosystem Issues, Compute Demand
News Content: At the 2024 annual seminar of the ChinaInfo100 (信息化百人会), a gathering of industry leaders, Professor Zheng Weimin, a member of the Chinese Academy of Engineering and a professor at the Department of Computer Science and Technology at Tsinghua University, delivered a thought-provoking speech. Professor Zheng highlighted the growing demand for computing power as AI model development shifts towards multi-modal applications, which are increasingly widespread across various sectors. This trend has led to an explosive increase in computing requirements.
However, compared to leading global enterprises like NVIDIA, domestic AI chips lag in the development of their system ecosystems. In his speech, Professor Zheng emphasized the critical importance of AI chip ecosystems. He argued that the maturity of the ecosystem directly influences the extent of chip adoption and usage, even if the performance is only 60% of what it could be. This underscores the urgent need for ecosystem development for domestic AI chips, underscoring the pivotal role of building a comprehensive, open, and interoperable ecosystem in driving the development of domestic AI chips.
Professor Zheng further analyzed that the multi-modal development of large AI models requires chips to process and optimize a variety of data types, including text, images, audio, and video. This demands higher levels of computational capability, memory management, and energy control from the chips. In contrast to international giants, domestic AI chips face challenges in ecosystem development, including deficiencies in software development tools, application development platforms, system integration capabilities, and industry solutions.
In response to this situation, Professor Zheng called for collective efforts from all sectors to accelerate the development of the domestic AI chip ecosystem. He stressed that ecosystem development requires the participation of governments, businesses, research institutions, and developers alike. By providing standardized interfaces, rich development tools, diverse application examples, and open platforms for communication, the ecosystem can facilitate rapid technology iteration and widespread application.
This speech not only illuminated the urgency of ecosystem development for domestic AI chips but also outlined a path forward for their future. By strengthening ecosystem development, domestic AI chips are poised to gain a competitive edge in the market, infusing new vitality into China’s artificial intelligence industry.
【来源】https://www.tmtpost.com/7160034.html
Views: 2