Online education, telemedicine, home office… The vigorous development of various new applications of digital economy has brought unimaginable changes to the whole society. Therefore, after the epidemic, China’s digital economy construction has obviously entered “acceleration”.
At the Fifth Plenary Session of the 19th CPC Central Committee, it was proposed to accelerate the development of digital economy, develop digital economy, promote digital industrialization and industrial digitization, promote the deep integration of digital economy and real economy, and build a digital industrial cluster with international competitiveness.
Artificial Intelligence Computing Center, a computing power “wormhole” that helps the development of digital economy
When the development of digital economy has become a national strategy, it is natural to need digital infrastructure that matches it. Under the upsurge of the fourth industrial revolution, industrial intelligence has promoted the vigorous development of AI industrial clusters, but what AI infrastructure needs to be matched to maximize the value of urban economies? Let’s start this space-time exploration together.
Strategic fortress of digital economy
The arrival of the intelligent era has brought a new industrial imagination to the regional economy.
The urban development path dominated by traditional economy needs congenital geographical conditions and geographical advantages, while the digital industry is not limited to physical location, but also depends on the cost advantages of electricity, rent, manpower and so on. The driving role for many urban economies is obvious. Among them, artificial intelligence, as one of the technical engines of the digital economy, has played a decisive role. McKinsey’s report mentioned that by 2025, AI could increase global GDP by 1.2% annually and China’s by 1.67%.
Artificial Intelligence Computing Center, a computing power “wormhole” that helps the development of digital economy
In recent years, the government departments of many cities have attached great importance to the AI industry. For example, Sichuan has released the Implementation Plan for the Development of New Generation AI in Sichuan Province (2018-2022) as early as 2018; Wuhan is also making efforts to build a national new generation AI innovation and development pilot zone. The East Lake High-tech Zone will establish a national leading talent training base for artificial intelligence industry; Xi’an, the largest capital city in the northwest, also issued the Action Plan for Xi’an to Build a National New Generation AI Innovation and Development Pilot Zone (2020 – 2022).
The fast development of the industry depends on the headband. The cities that plan to take the lead in the competition of digital economy have consciously deployed artificial intelligence computing centers to drive AI to process data in depth, support various intelligent services and applications, and make the cost of computing power better, the utility of intelligence greater, and better support the development of intelligent industry.
In this context, many cities have also begun to replicate their successful experience, and AI computing centers have naturally blossomed everywhere, becoming strategic fortresses to seize the digital economy highland.
So, what conditions do AI computing centers need to meet the needs of the digital economy?
First of all, the real needs of cities are the most critical coordinates. Many cities have artificial intelligence computing centers. On the one hand, they hope to obtain low-cost and high-performance AI computing power to support various intelligent applications in urban development; Secondly, infrastructure is needed to drive the convergence of AI industrial clusters, attract digital talents, and develop new innovation; The combination of the two can promote the healthy development of the regional digital economy. Therefore, in addition to the “infrastructure”, the AI computing center also needs to have a series of supporting solutions, ecological resources, talent training mechanism and long-term operation capabilities. We need to take a long-term view, do a good job of top-level design and strengthen the overall promotion, so that future development and actual progress can be “folded”, and build a path to the future for the city.
Open the space-time channel of urban intelligence
At present, Wuhan, Chengdu, Xi’an and other cities have started rapid preparation and operation together with technical partners to let AI run first.
Taking Wuhan as an example, the central computer room of the artificial intelligence computing center is constructed in a prefabricated modular way. It only takes 5 months from the construction to the official operation. On the day of putting into operation, the computing power load reached 90%, and then quickly entered the full load state, showing the “Wuhan speed” of rapid delivery of the project and the saturation operation of the project.
Why can these cities take the lead in opening the space-time channel of intelligent industry and seize the AI highland? Some commonalities can be found:
1. Talent is the engine.
Wuhan, Xi’an and Chengdu are provincial capital cities with numerous universities and scientific research institutes. The rich talent reserves are the most valuable resources in the development of AI industry, and can become the long-term driving force for the development of AI computing centers and industrial ecology.
Artificial Intelligence Computing Center, a computing power “wormhole” that helps the development of digital economy
2. Industry is vitality.
AI is a practical technology. Only AI computing centers that can truly serve the industry are qualified. The intelligent transformation of industrial enterprises needs to adapt to the AI model applied in the industry, and high-quality AI models are obtained through training and continuous iterative optimization. Therefore, the AI computing center needs to have the training ability of output algorithm model. The training chip should have floating point operation capability. In the training process of complex models, hundreds of billions of floating point parameters need to be fine-tuned by hundreds of thousands of steps, and fine floating point expression ability is required. If the manufacturer does not have a suitable training chip, it is difficult to guarantee the efficiency of the algorithm model produced by the platform, and the artificial intelligence computing center becomes a decoration.
It is reported that Wuhan Artificial Intelligence Computing Center has successfully assisted the Yangtze River Computing Assembly Workshop to realize the intelligent transformation of improving quality and efficiency, and the detection accuracy has been improved by 10%; The State Key Laboratory of Surveying, Mapping and Remote Sensing Discipline of Wuhan University has deeply cooperated with the Artificial Intelligence Computing Center to build a remote sensing image sample database (LuojiaSet) and a special remote sensing image framework (LuojiaNet), provide technology, platform and application support for major scientific research tasks such as natural resource monitoring, socio-economic development assessment, disaster emergency, help build China’s remote sensing scientific research ecosystem, and promote the industrialization of remote sensing applications in China, It really makes the AI computing center run and use. It can be seen that only a construction mechanism that closely combines AI computing power with application scenarios can ensure the long-term prosperity of the urban AI industry.
3. Self-control is the basic.
To hold the root firmly, with the intensification of the competition of national power and the decoupling of technology, high-rise buildings cannot be built on the sand dunes. In the long run, it is the only way to choose the fully stack of independent and controllable AI technology. But to be honest, at present, domestic AI technology still needs to make up lessons in the application ecological scale, which requires the joint efforts of urban decision-makers, software and hardware developers and technology enterprises.
To put it simply, AI infrastructure that allows talents to gather, apply, and control independently can form the soil for AI industrial clusters and truly open the “channel” for rapid urban development.
Supporting the ability shield of urban intelligence
On the other hand, how can autonomy and control be achieved? How to cultivate talents who adapt to the digital economy? How to incubate industrial AI innovation? Every step is no small challenge. After an in-depth observation of the exploration of several leading cities, we found that a user-friendly and usable AI computing center requires at least three prerequisites:
1. To provide sufficient computing power for the AI industry, we need to have a solid technical accumulation in the AI processor hardware. It is understood that at present, many AI computing centers use the training processor of the Zengteng 910 AI, which has the training ability natively, and the cluster performance is leading in the industry. At present, the cluster can complete training based on Resnet-50 model in 28 seconds (keeping the first in the industry), and its performance will continue to improve. Similarly, “Pengcheng Cloud Brain II”, based on the basic software and hardware platform of Shengteng AI, won the first place in AIPerf (world artificial intelligence computing power), and again refreshed the two world championships of IO500 (high performance computing storage system performance ranking list – full system input and output and 10-node system).
It is this surging computing power that can support accurate and reliable model training and reasoning, make AI innovation and application possible, and help the long-term sustainable development of the industry.
Artificial Intelligence Computing Center, a computing power “wormhole” that helps the development of digital economy
2. Enable the AI ecosystem of industrial cluster to form a virtuous circle. The AI computing center needs to provide comprehensive AI platform services with rich functions and the ability of multi-tenant and fine-grained resource management/isolation. It includes an integrated development environment that supports large-scale training data management and large-scale distributed training of large models. So that local governments, scientific research institutions, enterprises, colleges and universities can use computing resources quickly and safely.
Based on the basic software and hardware platform of Shengteng whole stack, the “Zidong Taichu” of the Institute of Automation of the Chinese Academy of Sciences has built a Chinese pre training model, a voice pre training model and a visual pre training model with excellent performance, and has innovatively realized the unified representation of vision text voice three modes through cross modal semantic association, and built a three mode pre training model, Give the cross-modal general AI platform multiple core capabilities.
3. In addition to the above conditions, artificial intelligence computing centers in Wuhan, Xi’an, Shenzhen and other places adopt the strategy of “getting on the horse and getting off the road”, introduce full-stack services including auxiliary operations, software infrastructure, hardware infrastructure, infrastructure, etc., and carry out overall long-term planning from the beginning, avoiding the gap between “building” and “using”.
Taking Xi’an Future AI Computing Center as an example, four ecological platforms (public computing service platform, application innovation incubation platform, industrial aggregation development platform, and scientific research and innovation talent training platform) have been built around one center to get through the “government-industry-learning-research-utilization”, so as to better play the value of the computing center service industry.
Artificial Intelligence Computing Center, a computing power “wormhole” that helps the development of digital economy
In the fourth wave of industrial revolution, artificial intelligence is undoubtedly the main line of technology. In order for cities to take the lead and occupy the AI highland, the first step is to “build high walls and accumulate grain widely”. The artificial intelligence computing center is the essential infrastructure and energy source.
Behind the rise of artificial intelligence computing centers is the time-space roadmap of China’s AI industry. Along this passage, the city and we are closer to the future.