Industry Insights | 'National Treasure': Research on China's Hypercomputing Service Market

Supercomputing, also known as high-performance computing, refers to the use of centralized computing resources from multiple computer systems working in parallel to handle extremely complex or data-intensive problems. Supercomputing possesses extremely powerful computing capabilities, with floating-point operation speeds reaching tens of billions per second. As one of the important strategic high points for national science and technology development, supercomputing is not only a reflection of a country's scientific and technological strength but also an important guarantee for the development of advanced high-precision technologies.

The supercomputing industry empowers data recovery and analysis, data simulation, and other scenarios by providing high-performance computing capabilities. From the perspective of service scenarios, the traditional supercomputing industry mainly focused on meteorology and oceanography, geological exploration, industrial simulation, physical simulation, etc., with service targets primarily including universities, scientific research institutes, and government-led technology projects. With the advent of the Internet, big data, and the Internet of Things era, more and more enterprises have an increasingly strong demand for data analysis. Facing the massive demand for data analysis and processing, general computers and servers can no longer meet the technical requirements, such as multimedia rendering, molecular model simulation, collision experiments, etc. In the future, empowered by technology, more and more enterprise service scenarios will face significant high-performance computing service demands. Supercomputing services will enter a new period of vigorous development driven by social economy and technological development.

Therefore, in terms of demand for computing resources, Frost & Sullivan categorizes high-performance computing into four main types: cutting-edge supercomputers, general-purpose supercomputers, business supercomputers, and artificial intelligence supercomputers.

• Before 2005: Exploration phase

With the state's support for high-performance computing during the '10th Five-Year Plan' period, Lenovo Group successfully developed the DeepTeng 6800 supercomputer system. The system's peak computing performance reached 5.3 trillion floating-point operations per second, leading the world at that time and far exceeding the special regulation target of 863.

• 2006 to 2015: Growth phase

During the growth phase, China has continuously accelerated the development of high-performance computers. In May 2010, the Shuguang 6000 was successfully developed, and in August 2010, the 'Tianhe-1' system was successfully developed. At that time, in the global supercomputer rankings, 'Tianhe-1' won the world title with a peak speed of 470 billion trillion times per second and a sustained speed of 256.6 billion trillion times per second. Subsequently, more supercomputers such as 'Shenwei BlueLight' and 'Tianhe-2' have been continuously introduced. By the end of 2015, the national high-performance computing environment included 15 node units with aggregated computing resources exceeding 12 PFLops and aggregated storage resources of 34 PB.

• 2016 to 2021: Maturity phase

During this period, our country has continuously accelerated the construction of supercomputing centers. Relying on self-controllable technologies, we are constantly building China's supercomputing ecosystem to provide users with more convenient and high-performance computing services. In addition, China also encourages various types of supercomputing centers to join the sharing of environmental computing resources to promote the generation of more industry application achievements. By the end of 2021, China currently had 9 national-level supercomputing centers located in Tianjin, Changsha, Jinan, Guangzhou, Wuxi, Shenzhen, Zhengzhou, Chengdu, and Kunshan.

• After 2021: Innovation phase

In the future, during the innovation phase, with the further development of 5G communication technology and the rapid application of emerging technologies such as artificial intelligence, cloud technology, big data, and blockchain, supercomputing services will integrate more emerging technologies and industries, achieving further upgrades in services. Supercomputing services integrated with emerging technologies will be able to serve more potential customers, especially enterprise customers and university clients, in a more flexible, powerful, and convenient mode.

On February 17, 2022, the National Development and Reform Commission, the Cyberspace Administration of China, the Ministry of Industry and Information Technology, and the National Energy Administration jointly issued a notice agreeing to initiate the construction of national computing power hub nodes in eight regions including Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, Chengdu-Chongqing, Inner Mongolia, Guizhou, Gansu, and Ningxia, and planning for 10 national data center clusters. The completion of the overall layout design of the national integrated big data center system marks the official launch of the 'East Data West Computing' project. In terms of specific construction goals, the 'East Data West Computing' project aims to orderly guide the demand for computing power in the east to the west by constructing an integrated new high-performance computing power network system including data centers and cloud services, optimize the layout of data center construction, and promote coordinated linkage between the east and the west.

From an industrial perspective, 'East Data, West Computing' has four major social missions: enhancing the overall computing power scale in China, promoting green industries, stimulating industrial investment and development, and coordinating regional common development to achieve shared prosperity.

• Overall computing power improvement

By constructing a nationwide integrated data center layout, we aim to expand the scale of computing power facilities, improve the efficiency of computing power utilization, and achieve large-scale and intensive development of computing power across the country.

• Promotion of green industries

The country will significantly increase the proportion of green energy used by data centers by expanding their layout in the western regions, localizing the consumption of green energy from the west. At the same time, through measures such as technological innovation, replacing large data centers with smaller ones, and low-carbon development, the country will continuously optimize the energy efficiency of data centers.

• Coordinate regional development and promote common prosperity

"The east data, west computing" strategy involves the layout of computing power facilities from east to west. This will drive the effective transfer of related industries, promote data circulation and value transmission between the east and west regions, expand the development space in the east, advance the formation of a new pattern for the large-scale development of the west, and accelerate the realization of China's social goal of common prosperity.

"The concept of 'East Data, West Computing' has a direct and positive impact on the hypercomputing industry, mainly reflected in further driving the development of third-party services and providing business support to independent hypercomputing service providers."

Currently, institutions of higher learning and research institutes in our country build internal micro- and small-scale high-performance computing environments on their own every year. Against this backdrop, the self-built model often suffers from the following pain points: First, the economic cost of self-building is significantly lower than using third-party professional services. Self-building requires a large amount of upfront investment and ongoing maintenance, which further leads to unnecessary use of funds; second, self-built supercomputing centers often result in waste of social resources. Compared to the national integrated layout and resource allocation model under the 'East Data, West Computing' model, self-built high-performance computing laboratories often do not possess high operational efficiency under intensive management, thus causing unnecessary waste of social resources. In the future, with the continuous construction of 'East Data, West Computing', the proportion of supercomputing service investments in China's self-built model will continue to decrease, while the proportion relying on professional third-party supercomputing services will significantly increase.

In the future, as users' recognition of third-party professional hypercomputing services increases, independent hypercomputing service providers, which prioritize computing performance, will become their preferred choice. Independent hypercomputing service providers will embrace new development opportunities.

• Geological exploration

Current energy geological exploration technology has entered the era of big data. To ensure the accuracy and efficiency of exploration results, geological exploration service providers usually need to analyze and simulate the physical structure and geological composition thousands of meters underground. In addition, with the popularization of 'wide azimuth, wide frequency band, high density' exploration technologies, each geological exploration measurement generates data volumes in the TB or even PB range. Facing such massive amounts of measurement data, further calculations and simulations are required to provide interpretable geological exploration results. High-performance computing has become one of the core solutions to improve geological exploration results.

• Meteorological, Oceanic Monitoring and Early Warning

High-performance computing can solve fluid motion states for a future period based on the physical characteristics of fluids, combined with scientific fluid modeling formulas including thermodynamics, fluid mechanics, atmospheric chemistry, and other dimensions. This enables monitoring and early warning effects. On the other hand, when obtaining fluid flow characteristics, it often requires multiple information acquisition channels such as radar observations and satellite observations, which brings about massive computational demands for information retrieval, analysis, processing, and modeling. In this scenario, only through supercomputing can the accuracy and timeliness of data processing be ensured.

• Bioinformatics and Life Sciences

Taking molecular model analysis as an example, bioinformatics modeling often involves massive molecular matching tests and corresponding model computations. Similar to this are gene sequencing, drug screening, and more. Faced with massive computational demands and data storage and invocation needs, only high-performance computing services can meet such needs.

• Rich media rendering

From a technical perspective, rich media rendering involves massive data acquisition, analysis, and processing. Moreover, typical rich media rendering scenarios have certain timeliness requirements, and the use of high-performance computing services is necessary to ensure that related business tasks can be delivered on time, thereby realizing maximum commercial value. Similar to bioinformatics, rich media rendering also favors hypercomputing services with high elasticity, because there are certain differences in the computational resource calls for each rendering. Enterprises often need to balance the relationship between business returns and production costs.

• Industrial simulation

Taking automotive machinery as an example, the simulation system can assist enterprises in aerodynamic analysis and prediction during model development, rigid strength analysis of vehicle body structures, collision analysis, etc. The simulation results can help engineers optimize and upgrade various processes and sub-projects. In the simulation system, it is often necessary to input a large number of design parameters, combined with intelligent algorithms, to achieve precise display of simulation results. Similar to automotive machinery, there are also manufacturing scenarios such as ship design and electronic chip simulation.

The 'main players' of hypercomputing services are still primarily national-led hypercomputing centers, aiming for ultimate computing performance. They provide sufficient computing power support for scientific research institutions, universities, and other entities. On the other hand, with the application of hypercomputing services in internet applications and industrial empowerment fields, independent hypercomputing cloud service providers and internet cloud hypercomputing service providers have actively entered the market. By balancing the ratio of machine computing power to commercial actual needs, they achieve the commercialization of hypercomputing services.

Compared with traditional supercomputing centers, independent hypercomputing cloud service providers and internet cloud hypercomputing service providers possess stronger customization capabilities and product design concepts, catering to the commercial needs of industry hypercomputing across a wide range of industries. They have actively promoted the exploration and model innovation of the hypercomputing industry by leveraging social forces.

Institute:Primarily focusing on cutting-edge supercomputing, the demand scenarios are mainly for innovative R&D and verification. The annual investment budget for supercomputing ranges from tens of thousands to hundreds of thousands of yuan, but there is a preference for hardware services, with high customization needs.

Higher education institutions:Primarily based on general-purpose supercomputers, the demand scenarios include engineering simulation, theoretical verification, research projects, etc. Leading institutions often invest more than one hundred million yuan in computing resources, mainly focusing on the procurement and upgrading of computing equipment, supplemented by cloud-based computing resources.

Corporate Clients:Primarily focused on high-performance computing for businesses, the demand scenarios include business optimization, R&D optimization, etc. Enterprises have very clear needs and purposes for supercomputing services, which are concentrated on optimizing business processes and costs to achieve commercial success. Enterprises usually have internal simulation research departments or computing optimization departments, and their demand for supercomputing services mainly focuses on resource requirements, with low customization requirements for the overall system.

In terms of market scale, due to the advantages of the charging model, enterprises have the greatest commercial demand for supercomputing. In 2021, this demand reached 121 billion yuan, accounting for more than 50%.

• The government continues to support the construction and development of the supercomputing industry.

Although China's supercomputing service market has experienced rapid growth and development, current supercomputing resources are still insufficient to meet the downstream demand for supercomputing services. With the continuous advancement of China's 'new infrastructure,' more and more university research laboratories, enterprises, and research institutes will face a stronger demand for high-performance computing services. To meet this explosive growth demand, the government will continue to increase the construction of supercomputing resources and the connectivity of ecosystems, providing a solid foundation for the interconnection of computing resources and services.

• The penetration rate of hypercomputing cloud continues to rise, and the market scale is growing rapidly.

Cloud services offer significant advantages such as rapid deployment, flexible scheduling, and elastic applications. As more researchers appreciate the convenience of cloud services, they hope to be freed from traditional self-built computing systems as soon as possible to achieve a higher-quality supercomputing service experience. In the context of supercomputing sub-scenarios, cloud services can greatly avoid the pain points of traditional self-built supercomputing centers, such as high capital investment, long procurement cycles, high maintenance difficulty, and slow resource updates. The penetration rate of supercomputing cloud will achieve 'dual-wheel drive' growth from two dimensions: the number of downstream customers and their business procurement volume.

• Container technology is becoming increasingly popular and has become a hot topic for the next generation of development.

Container technology is a technique that virtualizes and independently partitions running applications and their dependencies with lightweight operating systems. This ensures that processes within a container do not affect any processes outside of it. In past industry practices, container technology has been able to greatly improve R&D efficiency.

• Further integration of cluster technology and cloud service technology

Computer cluster technology has a wide range of applications in compute-intensive scenarios, such as atmospheric and oceanography, simulation modeling, geological exploration, etc. Cluster technology enables the expansion and utilization of computing power by interconnecting a group of independent servers through the network and managing them using cluster software systems. In the future, with the further penetration of industry supercomputing services, the close integration of cluster technology with cloud services will form a more flexible computing power configuration, bringing new innovation points to supercomputing cloud services.

• Data security processing capabilities continue to be strengthened

With the advent of the Internet of Things era, data security has become one of the focal points for various industries. How to protect one's digital assets and security information is the core demand of all users when choosing hypercomputing services. High-performance computing often involves the analysis and processing of large amounts of internal data. The stronger a hypercomputing service company has in data security management capabilities, the more it can bring business trust to its client enterprises, thereby achieving customer loyalty and commercial success.

• More mature industry-specific customization services

Supercomputing service scenarios such as marine meteorology, geological exploration, industrial simulation, and rich media rendering have varying requirements for supercomputing services, including the scheduling of supercomputing resources and the configuration of software systems. Only by having a profound understanding of each typical usage scenario can a product service matrix that conforms to the user habits of downstream users be designed.