The largest cloud based research and intelligent computing platform for Chinese universities has been launched, and the public cloud model has significantly improved the efficiency of computing power utilization

On June 27th, the largest cloud based research and intelligent computing platform CFFF (Computing for the Future at Fudan) in Chinese universities was officially launched at Fudan University. This scientific research "supercomputer" built to discover and solve complex scientific problems was jointly built by Fudan University and Alibaba Cloud

On June 27th, the largest cloud based research and intelligent computing platform CFFF (Computing for the Future at Fudan) in Chinese universities was officially launched at 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 thousand card parallel intelligent computing in an advanced public cloud mode, and supports the training of large models with hundreds of billions of parameters. This is the first example among domestic universities, and it is also ahead of Stanford University and other internationally renowned universities.

Continuing Fudan University's motto of "erudite and dedicated, asking and thinking", CFFF platform is composed of two parts: AIforScience intelligent computing cluster "Qienwen" No. 1 for multidisciplinary integration and innovation and special HPCC "Jinsi" No. 1 for high-accuracy and cutting-edge research.

Qi Yuan, professor Hao Qing of Fudan University and president of the Institute of Artificial Intelligence Innovation and Industry Research, introduced that based on the 100G high-speed data transmission network and Alibaba Cloud Cloud's globally leading large-scale heterogeneous computing power integration scheduling technology, hierarchical storage technology, AI and Big data integration technology, The "Jinsi" No. 1 deployed in Fudan University and the "Qiwen" No. 1 hosted in the Alibaba Cloud Ulanqab data center 1500 kilometers away have become a real "supercomputer". All experimental equipment in the four Fudan campuses can be accessed at high speed, achieving unified management of heterogeneous computing and unified scheduling of computing tasks to meet scientific intelligence research and application needs in different application scenarios.

The Alibaba Cloud Ulanqab data center, located on the "East Data and West Computing" node, provides intelligent computing services through the public cloud model of the Apsaras intelligent computing platform.

In 2022, Alibaba Cloud released the Apsara Intelligent Computing Platform and launched two large-scale intelligent computing centers, including the Ulanqab Data Center, to provide powerful intelligent computing services for scientific research, public services and enterprise institutions, which can increase the utilization of computing resources by more than three times, AI training efficiency by 11 times, and reasoning efficiency by six times.

It is reported that intelligent computing centers do not simply connect servers equipped with advanced process chips to achieve high-performance computing power. A crucial indicator is loss. Intelligent computing is even more challenging in this area. Compared with general-purpose computing, intelligent computing requires massive data for training. The loss of Data migration, distributed training and other links is especially serious. After the traditional intelligent computing center reaches a certain scale, increasing computing power resources will reduce the ability of computing power output. The minimum computing power output of the scale above 1000 calories is often only about 40%. However, Alibaba Cloud has significantly reduced the loss of intelligent computing power through systematic core technology self-development and taking the Apsara intelligent computing platform as the output.

Qi Yuan stated that relying on the public cloud model, projects running on the CFFF platform can enjoy the intelligent computing power of over a thousand card parallel, with the effective computing power of thousand card parallel reaching 92% of the industry's leading level, scalability reaching ten thousand card, and the effective computing power of ten thousand card parallel reaching 90%.

Meanwhile, public clouds are more low-carbon and green. Alibaba Cloud green data center technology combines with the natural climate advantages of Ulanqab. The CFFF platform can achieve an annual average PUE of less than 1.2, save more than 2000 kilowatts of total power annually, save 5 million yuan of electricity costs, and save 15 tons of carbon annually.

At present, the first scientific research achievement on the CFFF platform has been born. Li Hao's team from the Institute of Artificial Intelligence Innovation and Industry Research of Fudan University recently released a large model of medium and short term weather forecast with 4.5 billion parameters. The prediction effect reached the industry recognized collective average level of ECMWF (European Centre for Medium-Range Weather Forecasts) for the first time in the public data set, and the prediction speed was shortened from the original hourly scale to 3 seconds.

Based on the CFFF platform, a large-scale parallel intelligent computing model with thousands of cards can be trained in just one day. Traditional computing platforms are difficult to achieve, "said Li Hao.

This is also the first big model nurtured on the CFFF platform. Jin Li said that Fudan hopes to build a number of world-class scientific models based on the CFFF platform, such as life science model, material science model, Atmospheric science model, integrated circuit model, etc.

In the future, the CFFF platform will continue to expand its computing power scale and be open to research institutions, universities, hospitals, high-tech enterprises outside of Fudan University. On the same day, the first World Science Intelligence Competition for global scientific researchers was officially launched. The competition set up five major tracks, including life science and Quantum chemistry. CFFF platform will provide free training of computing power for participating teams, and provide long-term support for some scientific research projects with more inclusive computing power.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])