Chinese scientific excellence is developing quickly. Simultaneously, there is a rapid increase in supercomputing capacity. Combined, these may be indicative for the expanding diversification in the innovation landscape in China.
China is traditionally seen as a country with one dominant, central, top-down administration. As a capital, Beijing has been at China’s heart of governance, finance, science, education, industry and services for many centuries. In recent years, however, China has invested heavily in telecommunication, infrastructure and logistics, leading to an increased mobility of (human) resource and the development of expert hubs outside the traditional ones. Science and industry parks have started to compete for talent and other resources; and the “made in China 2025 strategy” demands enhanced level of knowledge generation and sharing between industries, which are traditionally uncomfortable with defining R&D agendas and outsourcing. Is it expected that Beijing will give up part of its intellectual hegemony by stimulating national competition, in a time that China’s government would like to stimulate the self-reliance of China? If so, High Performance Computing (HPC) may be one core technology that supports this development.
In December 2015, the Nature magazine published the Nature China index in which the Chinese high-quality scientific achievements over the last 2 years where discussed. The conclusion was unambiguous: “The Nature index shows that China is already a high-quality scientific powerhouse”. Advances made on the levels of institutes, cities, collaboration and industrial focus were highlighted. In summary, Beijing scored high on almost all levels. In terms of collaboration, however, Hong Kong, Hefei and Tianjin lead the pack; Wuhan and Shenzhen developed their strength in industrial focus; Nanjing excelled in academic output. An interdisciplinary Neuroscience and Technology institute in Hangzhou was also highlighted.
Chemistry and Physical sciences were topics that were most strongly represented. Government incentive lead to development of applications that serve industrial needs. On top of that, high-quality research outcome was also generated in the fields of Energy, Agriculture, Interdisciplinary (life) sciences, and Environmental sciences. Some of these topics are relatively new in the Chinese academic arena. Given the high quality research performance, it is expected that Chinese research organizations may take increasingly more initiative in international research and development programs.
The underlying framework for building up a diverse set of strong research capacities has co-occurred with a strong increase in HPC capacity in China. HPC may be seen as an enabler, providing dynamic platforms that can be adopted for many purposes, which are easy to access, and generate advanced data, products, insights, processes and/or tools. In the process of developing this technology in China, it has been a subject of improvement as well. Indeed, HPC in China indeed has awed the rest of the world. Within a decade or so, China has become a global HPC superpower. With respect to the numbers in the TOP500 (ranking the fastest HPCs globally), China is only second to the US (109 vs 199 in the end of 2015). The fastest HPC globally is the TianHe 2 in Guangzhou, and has been for 3 years consecutively. At present, operating very fast systems remains a challenge within the computer science community, and therefore TianHe 2’s predecessor, TianHe 1, is still more frequently used. Ca. 1000 users each day perform ca. 1400 tasks in animation, video effects, oil exploration data management, high-end equipment manufacturing and biomedical data processing.
The reasons for the rapid advance in Chinese HPC are grossly two-fold. First, this technology indeed is a key enabler. In non-military applications in China, HPC is used in scientific research, industrial design, aircraft design, drug discovery, animation, structural analysis, business support, simulations, diagnostic support, energy, telecommunications, weather, biotech, finance and more. Second, from a Chinese perspective, the need to develop a domestic computational value chain is real, as the United States has put restrictions on export licenses for HPC chips due to potential dual-use applications such as simulating nuclear reactions. The export of Intel Xeon chips to China from the US has been restricted from 2015. It is of interest to the Chinese, therefore, to accelerate domestic chip design and development even more. One successful example is the Sunwei (Shenwei) Bluelight chip, which has been introduced in the HPC infrastructure.
The geographical spread of Chinese HPCs is increasing. HPCs are located in the cities of Beijing, Tianjin, Shanghai and Guangzhou, Lanzhou, Dalian, Shenyang, and other cities. And more will be added. As software is not mature enough to efficiently operate the fastest systems, some regions that specialize in specific topics may benefit from more tailor-made HPC systems to serve their industry and communities. This is indicative for Chinese – and potentially global – trends. In line with this, establishing localized HPC systems may stimulate the knowledge-economy in regions outside China’s current main economic centers. Furthermore, data management and transfer/storage cost are becoming rate-limiting factors, instead of processor power. For instance, the world’s largest single dish radio telescope, which is being built in Guizhou, China (500 meter aperture spherical telescope, or FAST), will be served by its own tailor-made “Skeyeye-1” HPC. According to the South China Morning Post, the high cost of data transfer due to monopoly positions of state-owned telecommunication companies is one of the reasons for establishing the new Qiannan Supercomputing Center in Guizhou for Skyeye-1.
The HPC agenda in China has already spun of successful and internationally operating tech companies. Key industrial players are Sugon (Dawning) and Inspur. Sugon contributed to 45 novel HPC systems in a period of only 4 months last year in the TOP500. Inspur has existed for decades and is active in 85 countries. It has joint ventures and partnerships with Microsoft, LG, Ericsson, Intel, IBM, SAP, etc. While these commercial organizations, together with universities, generate open platform systems, their mission statements unambiguously express their support for developing “China’s information industry in the key stages of China’s informatization” (Inspur) or the “Data@China strategy leading to a connected environment, that meets governmental standards for an autonomous and controllable China” (Sugon). As China’s share in the global chip sales is about 50%, the main focus of these companies is indeed domestic at present, but future efforts may become more and more internationally oriented as well.
For HPC systems, the race for the speediest system is not over yet. Many US and EU systems are relatively old and will be upgraded soon. Suboptimal usage or overcapacity may slow down the Chinese growth over the next years. Also, the outcome of compliance restrictions between China and the US (after the US restricted export of chips, China has restricted export of systems) remains unclear. For now, it appears that China has been able to quickly ramp up its chip development, but the dependency of the US on systems from China is probably still limited. Finally, in line with Sugon’s mission statement (see above), government-imposed data distribution regulation might slow down the learning curve within the Chinese HPC domain. Whoever develops and operates HPCs most effectively has a powerful tool at hand: fast HPC systems now perform petascale (10^15) operations per second, while it is projected that exascale (10^18) operations per second can be realized in the beginning of next decennium. Then, petascale operations can be performed in smaller and more efficient devices, which is expected to be sufficient for AI-derived autonomy in drones and cars.
In conclusion, China is a scientific powerhouse that has a diversified innovation landscape. The increasing availability of advanced HPCs may continue to support this trend. For now, the distribution is still skewed towards Beijing and the other economic centers (Shanghai, Tianjin), both in terms of research diversification as well as HPC presence. With an ongoing increase in HPC systems, China possesses the in-house capacity to broaden its excellence in more fields and in more regions. This increases international competition and provides collaborative and/or supportive opportunities for innovation-driven companies abroad.
Sources, further reading