The national team enters the field and adds fuel to the intelligent computing ce

The national team enters the field and adds fuel to the intelligent computing ce

Recently, the AI artificial intelligence special promotion meeting held by the State-owned Assets Supervision and Administration Commission of the State Council (SASAC) has gone viral across the internet, attracting widespread attention from the industry.

The meeting called for central enterprises to actively embrace the profound changes brought about by artificial intelligence and to place the accelerated development of a new generation of artificial intelligence in a more prominent position. The meeting emphasized "to consolidate the foundation of development, concentrate the main resources into the fields that are most needed and have the most advantages, and accelerate the construction of a batch of intelligent computing power centers"; "to strengthen demand-driven, accelerate the empowerment of key industries, build a batch of high-quality multimodal data sets for the industry, and create a large model empowerment industry ecosystem from infrastructure, algorithmic tools, intelligent platforms to solution solutions."

At the meeting, 10 central enterprises signed an initiative, indicating that they will actively open up artificial intelligence application scenarios to society. As an important pillar of the national economy, the layout and development of central enterprises in the field of artificial intelligence are even more closely watched. The acceleration of the layout of artificial intelligence by central enterprises will bring new development opportunities to related industries, and also reflects the country's high attention and trend towards artificial intelligence.

Advertisement

With the rapid development of global AI technology, intelligent computing power has become a key element in promoting economic growth and social progress. As one of the world's largest economies, China is actively responding to this technological transformation, strengthening the research and development and application of intelligent computing power to enhance national competitiveness.

Before delving into China's layout in the smart computing market, let's first understand what "intelligent computing power" is? When do we come into contact with intelligent computing power in our daily lives?

01

What is intelligent computing power?

According to the definition in the "China Computing Power White Paper (2022)", computing power is mainly divided into four parts: general computing power, intelligent computing power, supercomputing power, and edge computing power. General computing power is mainly based on the computing capabilities output by CPU chips; intelligent computing power is mainly based on the artificial intelligence computing capabilities output by GPUs, FPGAs, AI chips, etc.; supercomputing power is mainly based on the computing capabilities output by supercomputers; edge computing power is mainly based on providing real-time computing capabilities to users nearby, and is a combination of the first three.From a conceptual standpoint, it may seem somewhat distant, but in reality, intelligent computing power has permeated every aspect of our lives. If we talk about the special effects, rendering, and facial recognition in Spring Festival movies, or the intelligent customer service and voice translation applications we use daily, all of these rely on the support of intelligent computing power.

If artificial intelligence is the accelerator of innovation, then smart computing centers can provide support for all kinds of technological innovations.

02

Large models create a gap in computing power, how much intelligent computing power does China need?

In 2023, numerous large model products were released. AIGC technology based on large models has shown effects in text generation, knowledge answering, image generation, and logical reasoning that far exceed expectations, attracting a large number of users and market attention.

According to incomplete statistics, as of October 2023, there were 254 large model innovation entities in China, distributed across more than 20 provinces/cities/regions, with Beijing leading the nation with 122. As of January 2024, 38 large models were registered and online, accounting for nearly half of the country. Baidu released the Wenxin Yiyan 4.0, with a user scale exceeding 100 million and a daily average of tens of millions of calls; ZhiPu AI developed the fourth-generation base large model GLM4, with the open-source version being downloaded more than 10 million times worldwide, making it the most influential open-source domestic large model at present; Baichuan Intelligence released the large model Baichuan2, which achieves the longest context window globally, with the open-source version being downloaded more than 6 million times worldwide in just four months; The Institute of Automation of the Chinese Academy of Sciences released the world's first trillion-parameter multimodal large model, Zidong Taichu 2.0. In terms of deep learning frameworks, Baidu's PaddlePaddle is in the first echelon of domestic popularity and usage, with a domestic market share of nearly 36%. As of the end of December 2023, it has gathered 10.7 million developers and served 235,000 enterprises and institutions.

In practical applications, the production method of "digital humans" created by using AI technology to automatically generate content rivals the level of real people; artificial intelligence prediction of protein structures brings a new research method to basic research; AI-driven chatbots can learn and understand human language and converse with humans; Huawei Cloud's "Pangu Meteorological Large Model" has shown its prowess in predicting typhoon trajectories and landing times...

Behind the thriving development of visible AIGC is the invisible support of intelligent computing power. The explosion of large models has ignited a new round of AI fervor and has also changed the demand and landscape of intelligent computing power.

The "Action Plan for High-Quality Development of Computing Power Infrastructure," jointly issued by the Ministry of Industry and Information Technology, the Central Cyberspace Affairs Office, the Ministry of Education, the National Health Commission, the People's Bank of China, and the State-owned Assets Supervision and Administration Commission, proposes that by 2025, China's computing power scale will exceed 300 EFLOPS, with intelligent computing power accounting for 35%.Speaking of this, some may wonder: What is FLOPS? What level is 300EFLOPS?

FLOPS is a unit of computing power, measuring the number of floating-point operations that computing resources can perform per second, which is the abbreviation for Floating-point operations per second. It is often used to estimate the execution performance of computers, especially in the field of scientific computing that involves a large number of floating-point operations, such as training and inference of image processing related to machine vision.

The letters in front of FLOPS represent larger units of computing power:

- One MFLOPS (megaFLOPS) equals one million (=10^6) floating-point operations per second.

- One GFLOPS (gigaFLOPS) equals one billion (=10^9) floating-point operations per second.

- One TFLOPS (teraFLOPS) equals one trillion (=10^12) floating-point operations per second.

- One PFLOPS (petaFLOPS) equals one quadrillion (=10^15) floating-point operations per second.

- One EFLOPS (exaFLOPS) equals one hundred quintillion (=10^18) floating-point operations per second.

Here are a few examples for better understanding: When training artificial intelligence models with a large amount of data samples, the computing power required for one training session varies from 2-19PFLOPS depending on the data scale, detection effect, and model category; during the inference process of face and language recognition, the demand for computing power may range from 10GFLOPS to 64TFLOPS depending on the recognition accuracy and the number of concurrent operations; for intelligent driving to complete functions such as environmental perception, decision-making for obstacle avoidance, and self-vehicle positioning, the computing power requirement is approximately 8TFLOPS.

As of the end of June 2023, the total scale of data center racks in use nationwide exceeded 7.6 million standard racks, with a total computing power of 197EFLOPS, ranking second in the world.Driven by the demand for large-scale models, intelligent computing center projects have emerged like mushrooms after a spring rain.

03

Over 30 cities are racing to build intelligent computing centers

With the concentrated outbreak of downstream computing power demand and the advancement of the "East Data West Computing" initiative, governments at all levels, operators, and internet companies have all launched plans to build intelligent computing centers. According to the "Guidelines for the Innovative Development of Intelligent Computing Centers" jointly published by the National Information Center and relevant departments, more than 30 cities across the country are currently building or proposing to build intelligent computing centers, with classic cases including the Beijing-Tianjin-Hebei Big Data Intelligent Computing Center and the Changsha 5A-level Intelligent Computing Center.

The main entities constructing intelligent computing centers include the three major telecommunications operators and some internet companies. The intelligent computing centers promoted by operators have a certain public service attribute, becoming a good supplement to the government-led computing power infrastructure construction. Internet companies represented by Baidu, Alibaba, and Tencent are also building intelligent computing centers to promote their own business development and better facilitate the implementation of artificial intelligence scenarios for their customers.

According to statistics from the Forward Intelligence Center, as of August 2023, China has intelligent computing centers in operation and under construction in cities such as Beijing, Shanghai, Nanjing, and Hangzhou. Looking at the regional distribution, Chinese intelligent computing centers are concentrated in the eastern and central regions. Among them, the eastern region has 25 intelligent computing centers, accounting for 62.5%, ranking first, mainly in the Beijing-Tianjin-Hebei and Yangtze River Delta areas; the central region accounts for 17.5%, ranking second; the western and northeastern regions account for 12.5% and 7.5% of the intelligent computing centers, respectively.

It is worth noting that Beijing is one of the main regions focusing on intelligent computing centers. Beijing has laid out intelligent computing centers in various districts such as Haidian, Chaoyang, Economic Development Zone, Shijingshan, Mentougou, Daxing, and Fengtai, with a total scale of intelligent computing power of about 5000P currently built.

Recently, the four-span factory building in the BeiZhong Science and Technology Cultural Industry Park in Shijingshan District has started a hot renovation construction. After completion, it will initially have a computing power of 610P, equivalent to the computing capability of 300,000 high-performance computers, which can support an artificial intelligence large model to complete the learning and recognition of nearly 10 million images within 30 seconds. It is expected to be completed and put into use in October this year.

The year has just begun, and there are many actions.At the start of the New Year, major operators have actively taken action, strengthening their layout in the field of intelligent computing centers.

On January 8th, China Mobile's Intelligent Computing Center (Wuhan) commenced operations in Wuhan's Future Science City, having built a service capability of 1500 PFLOPS, with plans to expand to 6800 PFLOPS by the end of this year, becoming the largest intelligent computing center in the Central China region.

On January 22nd, Shanghai Telecom illuminated the "Large-scale Computing Cluster and Artificial Intelligence Public Computing Service Platform" in Shanghai, planning to build a total of 15,000 cards in Shanghai by the first half of 2024, with a total computing power exceeding 4500 P, of which a single pool will newly build a domestic computing power of 10,000 cards, expected to become the first ultra-large scale domestic computing power liquid-cooled cluster in the country.

On January 30th, the establishment ceremony of China Unicom's Artificial Intelligence Innovation Center was held in Beijing. It is worth noting that only on November 24, 2023, did the China Unicom Yangtze River Delta (Wuhu) Intelligent Computing Center project officially commence.

It can be seen that the three major telecommunications operators, Telecom, Mobile, and Unicom, are focusing on the construction of the "East Data West Computing" data center cluster, fully promoting their respective related projects, accelerating the creation of national computing power center cities and intelligent computing centers, and promoting the deep integration of the digital economy with the real economy.

04

AI servers are the most critical equipment in the construction of intelligent computing centers.

According to the latest report from market research institution IDC, from the first half of 2022 to the first half of 2023, the scale of China's AI server market grew by 54%, with GPU servers still holding a dominant position, accounting for 92% of the market share, reaching 3 billion US dollars. At the same time, non-GPU accelerated servers such as NPU, ASIC, and FPGA grew at a year-on-year rate of 17%, accounting for 8% of the market share, reaching 200 million US dollars.In the first half of 2023, from the perspective of manufacturer sales revenue, Inspur, H3C, and Ning Chang ranked in the top three, accounting for more than 70% of the market share; from the perspective of server shipment units, Inspur, Kunqian, and Ning Chang were the top three, holding nearly 60% of the market share.

AI servers rely on the supply of high-performance chips. The computational power gap faced by the Chinese market brings new opportunities for the development of domestic chips. Chinese domestic AI chip manufacturers are in a stage of rapid growth and have achieved significant accomplishments, attracting a large amount of investment and attention. These companies have certain strengths and competitive advantages in AI chip design, algorithm optimization, and manufacturing. In addition, the policy support from the Chinese government has played an important role in promoting development. In the first half of 2023, the market size of China's accelerated chips exceeded 500,000 units. From a technical perspective, GPU cards hold 90% of the market share; from a brand perspective, domestic Chinese AI chip brands shipped more than 50,000 units, accounting for about 10% of the entire market share.

Looking at the suppliers of AI acceleration chips required for domestic AI servers, domestic cloud service providers such as Alibaba (including the Guang series), Baidu (Kunlun series), and Huawei (Ascend series) have developed their own cloud AI acceleration chips. There are also Cambricon (Xuanyan series), Sugon Information (Deep Computing series), Suiyuan Technology, TianShu Zhixin, BiRan Technology, Moore Threads, and Muxin, among others. In addition, Jingjia Micro and Loongson Technology are also developing cloud AI acceleration chips.

According to data from the first half of 2023, the domestication rate of Chinese AI server chips has declined compared to last year, dropping from about 15% to about 10%. This is mainly due to the significant increase in demand for high-end training servers, while the performance of domestic chips is difficult to keep up.

To further enhance the performance of domestic AI servers, it is necessary not only for chip manufacturers to continue technological innovation and improve the performance and stability of chips but also for these chip manufacturers to deeply understand market demands and develop chip products that are more in line with actual application scenarios. At the same time, the government, enterprises, and research institutions should also increase investment to support the development of the domestic chip industry, providing more research and development resources and market opportunities.

05

How to make the smart computing center truly operational?

After the completion of the smart computing center, how to make it play a greater role in the operation process is still a very critical issue.

Currently, smart computing centers have gradually empowered the development of regional industrial clusters, but it is worth noting that they still face many challenges in terms of multi-computational power integration, upstream and downstream collaboration, construction and application linkage, energy consumption, and usage pricing.For instance, the integration of general-purpose computing power and specialized computing power is still pending. In different scenarios such as autonomous driving, smart healthcare, and smart cities, the demand for computing power varies. A one-size-fits-all computing power solution struggles to meet diverse computing power needs, failing to cater to multiple industries and fields.

There is insufficient synergy between computing power, algorithms, and data. Over the years, the intelligent computing centers that have been built have seen different chip platforms, algorithmic models, databases, and application layers in a state of vertical integration "islands," with software and hardware compatibility issues that need to be improved.

Investment, construction, and operation need to be better coordinated. The investment, construction, and operation of intelligent computing centers are often managed by different entities. The construction units in the early stages often do not invest enough in the operational models and service standards after construction, leading to a phenomenon of managing the beginning but not the end, and a disconnect between construction and operation, which affects customer experience.

High carbon emissions and energy consumption are a significant challenge. The energy consumption and emissions of the equipment itself pose a considerable challenge, such as the electricity consumption required for training the ultra-large-scale pre-trained model GPT-3 by OpenAI, which is 190,000 kilowatt-hours, equivalent to 228 times the per capita electricity consumption in 2021.

Investment costs and application pricing need to be standardized. The cost of building intelligent computing centers is relatively high, with some centers costing up to 500 million to 600 million yuan for every 100P of semi-precision computing power, far exceeding the normal market price. Additionally, the usage cost is also high, for example, it is conservatively estimated that the training cost for the GPT-3 large model exceeds 12 million US dollars.

The construction of intelligent computing centers needs to be combined with the construction foundation and local or regional industrial characteristics, with categorized guidance and policy implementation, renovation and parallel development, and the growth of intelligent computing centers adapted to the digital economy. It is also necessary to choose reasonable construction and operation models to achieve long-term operation, promote orderly layout, and ensure the maximization of the economic and social benefits released by intelligent computing centers.

Currently, the development of China's intelligent computing center industry is overcoming the challenges of the 1.0 era and entering the 2.0 era. In the construction of intelligent computing centers, China has always adhered to the principles of computing power integration, software and hardware collaboration, integrated construction and operation, low-carbon energy consumption, cost optimization, demand-driven, and security and trustworthiness, steadily promoting the development of intelligent computing centers.It seems like there is no text provided for translation. Please provide the text you would like translated into English, and I will be happy to assist you.