On June 27th, it was announced that the largest cloud-based scientific research and intelligent computing platform for universities in China, CFFF (Computing for the Future at Fudan), officially launched at Fudan University.
This “supercomputer” for discovering and solving complex scientific problems was jointly created by Fudan University and Alibaba Cloud, providing advanced public cloud-based parallel intelligent computing with over a thousand calories to support large-scale model training with billions of parameters. This is the first case among Chinese universities and also ahead of internationally renowned universities such as Stanford.
According to the introduction, the CFFF platform consists of two parts: “Qie Wen” No.1, an AI for Science intelligent computing cluster that focuses on multidisciplinary integration innovation, and “Jin Si” No.1, a dedicated high-performance computing cluster for advanced precision research.
According to the introduction, based on a 100 Gbps high-speed data transmission network and Alibaba Cloud’s globally leading large-scale heterogeneous computing fusion scheduling technology, hierarchical storage technology, AI and big data integration technology, “Jin Si” No.1 deployed on Fudan campus and “Qie Wen” No.1 hosted in Alibaba Cloud’s Ulanqab data center 1500 kilometers away have been connected into a truly meaningful “supercomputer”.
All experimental equipment in Fudan’s four campuses can access it at high speed, achieving unified management of heterogeneous computing power and unified scheduling of computing tasks to meet scientific intelligent research and application needs under different scenarios.
Currently, the first scientific research achievement on the CFFF platform has been born. Recently, Li Hao’s team from Fudan University’s Artificial Intelligence Innovation and Industry Research Institute released a large-scale model for medium-term weather forecasting with 4.5 billion parameters. The prediction effect reached the industry-recognized ECMWF (European Center for Medium-range Weather Forecasts) ensemble average level on public datasets for the first time, and reduced the prediction speed from hours to within 3 seconds.
The intelligent computing platform represented by the CFFF platform, as an emerging scientific supercomputing architecture, will become an important support force for scientific research, greatly improving research efficiency and reducing research costs. It will accelerate the discovery of scientific principles and technological breakthroughs, and effectively promote the implementation of large-scale scientific models.