In recent years, driven by technology research and development and industrial applications, China's supercomputers (referred to as "supercomputers") have developed rapidly. In terms of technological innovation, many domestic supercomputers using self-developed chips have been on the world's supercomputing list. Ranked first; in terms of overall scale, China accounts for more than 1/3 in the latest global supercomputing list; in terms of application innovation, since 2016, multiple innovative applications have won the highest award in the world of high-performance applications, the "Gordon· Bell" award.
Supercomputing has always been the country's most important high-performance computing technology in the field of computing infrastructure, which to a large extent reflects the country's comprehensive scientific research strength. Massive data processing capabilities and high-density computing capabilities are intuitive standards for measuring the performance of supercomputers. The industry usually uses computing power (floating point operation speed) and computing efficiency (floating point operation speed per watt) to measure the absolute performance and energy efficiency of supercomputers. Make an evaluation. Ten years ago, the absolute computing power of international supercomputers had reached P level, that is, it could perform quadrillion mathematical operations per second.
High-performance computers (HPC, also known as supercomputers) are mainly used to solve challenging problems that cannot be solved by other computers. They usually refer to a type of computer with extremely fast computing speed, huge storage capacity, and extremely high communication bandwidth. Supercomputing is regarded as the "crown" of computer science and engineering, and countries have frequently launched research and development plans at the national level in recent years.
Supercomputers can perform numerical simulations and dynamic displays on the objects under study, thereby obtaining results that are difficult or even impossible to obtain experimentally. For example, it can be used in fields such as predicting climate change, forecasting severe weather, understanding gene function, simulating nuclear explosion processes, and even simulating the evolution of the universe. The development of supercomputers has made computational science a new method and approach for humans to understand and transform the world.
Supercomputing is the “crown” of computer science and engineering. Supercomputing is one of the sources of computer technology. The core technologies of data centers that the Internet industry relies on, such as parallel programming tools such as Hadoop and remote communication technologies such as RDMA, are mostly derived from this. Therefore, supercomputers are regarded as the "crown" of computer science and engineering.
Continuously developing supercomputing technology can support technical research in major scientific fields and promote application innovation in emerging scientific and technological fields, playing a key role in leading national scientific and technological progress. Over the years, supercomputing technology has been widely used in major scientific research and national economy and people's livelihood fields such as aerospace, electromagnetic computing, ocean modeling, medical imaging, and materials genetics, making important contributions to the country's basic scientific research and technological research in related fields. As more real-world applications move from data-based simulation and emulation to algorithm-based mining and training, emerging technologies such as big data and artificial intelligence are integrating with supercomputing technology. Supercomputing has gradually become the computing infrastructure that supports the development of informatization and the computing brain of cities and key industries.
After decades of development, my country's supercomputing has made certain achievements, but it still faces many challenges. First, the adaptation problems caused by technological heterogeneity are gradually emerging. Different chip architectures ensure supercomputing implementation path redundancy, but with the in-depth development of various applications, they also bring about software adaptation problems. Secondary adaptation is required between models, algorithms and chips, making application software transplantation difficult to achieve. Second, the impact brought by intelligent computing centers is becoming increasingly severe.
Since 2020, the 14th Five-Year Plan and the new infrastructure have driven the construction of supercomputing centers in our country into a period of rapid growth. Many local governments, enterprises and institutions are actively building and preparing for the establishment of supercomputing centers.
After China has repeatedly been listed in the global supercomputing TOP500, the United States has included a number of Chinese supercomputing-related institutions or enterprises on the entity list since 2015, including the National University of Defense Technology, Wuxi Jiangnan Computing Technology Research Institute, Sugon, Shenwei, etc. However, there is still a large gap between my country's overall supercomputing strength and that of the United States.
Judging from the performance share of Linpack in the industry application field among China's TOP100 supercomputers in November 2021, computing power services, supercomputing centers, artificial intelligence, scientific computing and other fields are the main users of supercomputing. Internet big data, especially the field of AI, is growing. Strong.
The 14th Five-Year Plan and new infrastructure have driven supercomputing into a period of rapid growth. The construction of high-performance computing centers in my country mainly establishes the construction plan of supercomputing centers through cooperation agreements between ministries and provinces (municipalities). The Ministry of Science and Technology sets goals for host performance on behalf of the national science and technology strategy. The local government hopes that the supercomputing center can become a functional carrier for regional scientific and technological development, gathering talents, innovating science and technology, and promoting economic development.
In March 2021, my country's "14th Five-Year Plan" clearly stated that it is necessary to "accelerate the construction of a national integrated big data center system AZ, strengthen the overall intelligent dispatching of computing power, build a number of national hub nodes and big data center clusters, and build E-class and 10E-level supercomputing centers.” According to the plan, supercomputing centers will be established in Hefei, Lanzhou, Xiamen, Taiyuan and other places.
In recent years, the United States, the European Union, Japan and other countries and regions are accelerating the realization and optimization of E-class computing performance goals (which can perform tens of billions of mathematical operations per second). China also attaches great importance to the development and research of E-class supercomputers and has successfully developed multiple E-class prototypes. Compared with foreign countries, the characteristics and advantages of China's supercomputers are mainly reflected in aspects such as computing performance, independent research and development, and green energy saving.
With the rise of artificial intelligence demand, more than 20 intelligent computing centers have been built across the country since 2019, which has had a great impact on the development of supercomputing. How to achieve differentiated development of supercomputing and intelligent computing centers is an important issue facing the industry. Third, the challenges brought by the business model have gradually become a reality. After the country's supportive policies for supercomputing are weakened, how supercomputing centers can find a sustainable development path that is responsible for their own profits and losses, or even a slight surplus, is a practical problem facing operators.