Credit: Visual China
【资料图】
BEIJING, June 28 (TiPost) – Fudan University announced on Tuesday the official launch of CFFF (Computing for the Future at Fudan), a cloud-based intelligent computing platform developed in collaboration with Alibaba Cloud and China Telecom.
The platform boasts an impressive overall computing power of 28 TFLOPS (trillion floating-point operations per second) and is divided into two components. The first is a dedicated high-performance computing cluster named “Jinsi” No. 1, which is deployed on the university campus, primarily utilized for advanced research. The second component is “Qiewen” No. 1, which is hosted in Alibaba Cloud’s Ulanqab data center. This cluster enables parallel computing across more than 1000 NVIDIA GPU boards, facilitating the training of large models with up to 100 billion parameters.
China Telecom is responsible for constructing a 100G high-speed data transmission network. This network enables seamless and high-speed access to the CFFF platform for all experimental equipment across the university’s four campuses, facilitating unified scheduling and efficient data exchange.
Jin Li, the president of Fudan University, emphasized the transformative impact of scientific intelligence (AI for Science) on the research paradigm. “The paradigm shift focuses on leveraging vast amounts of scientific big data and employing new technologies such as multimodal approaches and large models to discover new substances, synthesize new materials, and develop new mechanisms,” he said. With the CFFF platform as its foundation, the university aims to develop several world-class scientific big models. These models will span various disciplines, including life sciences, material sciences, atmospheric sciences, integrated circuits, and more.
Jin highlighted that currently, only around 50 to 60 research groups within the university are engaged in AI-related work, accounting for just approximately 1% of the university’s total student population. He expressed his vision of developing AI training courses for the platform, with the aim to equip most students with not only an understanding of AI but also the ability to innovate with AI.
In its initial phase, the CFFF platform will primarily cater to the research requirements of various departments at Fudan University. It will support research in fields such as life sciences, atmospheric sciences, material sciences, as well as social science areas like financial system analysis.
Recently, Li Hao’s team at the university’s Institute of Artificial Intelligence Innovation and Industry utilized the platform to develop a large-scale model for short- and medium-term weather forecasting. The model, consisting of 4.5 billion parameters, achieved prediction results that reached the industry-recognized ECMWF (European Center for Medium-Range Weather Forecasts) aggregate average level for the first time on the public dataset. Furthermore, the model significantly improved the prediction speed from hourly intervals to within just 3 seconds.
In the future, the CFFF platform aims to expand its computing power and gradually open access to research institutions, universities, hospitals, high-tech enterprises, and other entities beyond the university.
Wang Jian, an academician of the Chinese Academy of Engineering and the founder of Alibaba Cloud, shared that it took approximately six months to build the platform. Alibaba Cloud dedicated an engineering team of more than a hundred people to develop its structure free of charge.
Wang expressed concerns about the current business challenges associated with big models and highlighted the potential risks if they are solely driven by capital. He emphasized that involving educational institutions in the development of big models can contribute to the sustainability of this field.
关键词：
