Shulou(Shulou.com)11/24 Report--

CTOnews.com June 27 news, today's domestic universities largest cloud scientific research intelligent computing platform CFFF (Computing for the Future at Fudan) officially launched in Fudan University. This scientific research "supercomputer" built to discover and solve complex scientific problems was jointly built by Fudan University and Alibaba Cloud. It provides ultra-kilocalorie parallel intelligent computing in advanced public cloud mode and supports large model training with 100 billion parameters. This is the first time in domestic universities and is also ahead of internationally renowned universities such as Stanford University.

According to introduction, CFFF platform consists of two parts: AI for Science intelligent computing cluster "Inquiry" No.1 oriented to multidisciplinary integration innovation and special high-performance computing cluster "Jinsi" No.1 oriented to high-precision research.

Professor Haoqing of Fudan University and President of the Institute of Artificial Intelligence Innovation and Industry, Qi Yuan, introduced that based on the 100 G high-speed data transmission network and Alibaba Cloud's world-leading large-scale heterogeneous computing power fusion scheduling technology, hierarchical storage technology, AI and big data integration technology, the "Jinsi" No.1 deployed in Fudan University and the "Qiewen" No.1 hosted in Alibaba Cloud Wulanchabu Data Center 1500 kilometers away have been connected into a real "supercomputer". All experimental equipment in Fudan No.4 Campus can be accessed at high speed, so as to achieve unified management of heterogeneous computing power and unified scheduling of computing tasks, so as to meet the scientific intelligent research and application requirements under different application scenarios.

According to Qi Yuan, depending on the public cloud mode, projects running on the CFFF platform can enjoy ultra-kilocalorie parallel intelligent computing power, the effective computing power of kilocalorie parallel reaches 92% of the industry leading, the scalability reaches 10,000 calories, and the effective computing power of 10,000 calories can also reach 90%.

At the same time, public clouds are more low-carbon and green. Alibaba Cloud Green Data Center technology combined with the local natural climate advantages of Ulanqab, CFFF platform can achieve an annual average PUE of less than 1.2, save more than 2000 MWh of total electricity every year, save 5 million yuan of electricity, and save 15 tons of carbon annually.

At present, the first scientific research achievement on CFFF platform has been born. CTOnews.com notes that Li Hao's team from the Institute of Artificial Intelligence Innovation and Industry of Fudan University recently released a large medium-and short-term weather forecast model with 4.5 billion parameters. For the first time, the prediction effect reached the industry-recognized ECMWF (European Center for Medium-Term Weather Forecasting) aggregate average level on the public dataset, and shortened the prediction speed from the original hourly level to less than 3 seconds.

"Based on the Kcal parallel intelligent calculation of CFFF platform, such a large-scale model was trained in only one day. Traditional computing platforms are hard to do. "Li Hao said.

Jin Li, president of Fudan University and academician of Chinese Academy of Sciences, said that in the "big science era" driven by data and intelligent technology, how to win the initiative in the ever-changing scientific and technological innovation environment and make innovation breakthroughs in key fields is the proposition given to colleges and universities by the times. As a new scientific research supercomputing architecture, the intelligent computing platform represented by CFFF platform will become an important supporting force for scientific research, greatly improving scientific research efficiency, reducing scientific research costs, accelerating scientific principle discovery and technological breakthrough, and strongly promoting the landing of scientific large models.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.