Author: Gold Leaf
Behind the unabated popularity of ChatGPT is the increasing demand for computing power required by the automatic content generation technology of artificial intelligence.
The data shows that the total computing power consumption of ChatGPT is about 3640PF-days (that is, if the calculation is 1 quadrillion times per second, it needs to be calculated for 3640 days), and it needs 7~8 data centers with an investment scale of 3 billion and a computing power of 500P to support the operation .
So what is the current computing power situation in the world and in my country?
The demand for computing power is huge
Computing power is the core strength to promote the development of the digital economy and the solid foundation to support the development of the digital economy.
In the context of the digital economy era characterized by all-things perception, all-things interconnection, and all-things intelligence, the total amount of global data and the scale of computing power continue to show a trend of rapid growth.
China Academy of Information and Communications Technology estimates that every 1 yuan invested in computing power will drive 3 to 4 yuan of economic output.
In 2021, the scale of my country's computing power industry will reach 2.6 trillion yuan, directly and indirectly driving the total economic output of 2.2 trillion yuan and 8.2 trillion yuan respectively.
"Computing power, algorithms, and data, as the three elements of artificial intelligence, play a huge role in the development of the digital economy and the upgrading of industrial intelligence. Data can be regarded as means of production, algorithms represent new production relations, and computing power is a new type of productivity. Algorithms and data are supported, and the level of computing power determines the strength of data processing capabilities. In the process of AI model training and reasoning operations, strong computing power support is required. With the increase in training intensity and computational complexity, the requirements for computing power accuracy also increase. It is gradually improving.” said Zou Debao, executive deputy general manager of CCID Consulting’s Artificial Intelligence Industry Research Center.
According to the National Data Resources Survey Report, the total global data output will be 67ZB in 2021, with an average growth rate of more than 26% in the past three years.
Basic computing power, intelligent computing power, and super computing power all show a steady growth trend.
"Since the advent of AlphaGo in 2016, the demand for intelligent computing power has started to explode. Now ChatGPT represents a new round of explosion in demand for AI computing power." said He Renlong, chief expert of Shanghai Emerging Information and Communication Technology Application Research Institute.
He explained to the first financial reporter that ChatGPT is an AI natural language model algorithm in terms of algorithm type, which belongs to the "alpha dog" of NLP (natural language processing).
"The Go version of AlphaGo is self-calculation with limited data, while ChatGPT is open data. Internet data all over the world is in the computing category, so the amount of data is infinitely larger than that of AlphaGo, and the computing power consumption is also greater than that of search engines (indexing technology). Several times, and it is still GPU computing power."
The scale of computing power in each country is closely related to the level of economic development. The higher the level of economic development, the larger the scale of computing power.
For ChatGPT, a "gold-swallowing monster" that consumes a huge amount of computing power, infrastructure data center investment is the foundation.
According to the "2021 Global Artificial Intelligence Innovation Index Report" previously released by the Institute of Scientific and Technical Information of China, the data center retention rate and the proportion of the global TOP500 supercomputing centers are regarded as the main reasons for the differences in the basic support scores of the innovation index of various countries.
In terms of this subdivision indicator, the retention rate of data centers in the United States has reached a full score, and China has a clear advantage in the proportion of TOP500 supercomputing centers in the world, and the number has maintained the first place year after year.
As of the end of 2021, the total scale of data center racks in use in my country will reach 5.2 million standard racks, the scale of data center servers in use will be 19 million, and the total scale of computing power will exceed 140 EFLOPS.
There are more than 450 ultra-large and large data centers in use nationwide, and more than 20 intelligent computing centers.
Although the number of supercomputing centers in China that entered the top 500 in the world from 2018 to 2021 has always maintained a leading position in the world, it is worth noting that as more and more countries begin to increase the construction of computing power infrastructure, compared with 2020 , China's advantage in supercomputing centers has weakened.
"From the perspective of the number of supercomputing centers, China has surpassed the United States, but there is still a gap in the computing foundation. The core is mainly high-quality data centers that need to be improved." Xu Xu, executive deputy director of the Artificial Intelligence Development Research Center of the China Institute of Scientific and Technological Information Feng told China Business News.
Intelligent computing power should be strengthened
From the perspective of industry insiders, ChatGPT's requirements for computing power mainly focus on intelligent computing power, which is also an aspect that our country needs to strengthen urgently.
The "White Paper on China's Computing Power Development Index (2021)" shows that the proportion of my country's smart computing power has increased from 3% in 2016 to 41% in 2020.
During the "13th Five-Year Plan" period, my country's general-purpose computing power has increased by three times, and the intelligent computing power has increased by nearly a hundred times.
According to the "2022-2023 China Artificial Intelligence Computing Power Development Evaluation Report", the scale of China's intelligent computing power will reach 155.2 EFLOPS (FP16) in 2021, and it is expected that by 2026, the scale of China's intelligent computing power will reach 1271.4 EFLOPS.
During the period from 2021 to 2026, the compound annual growth rate of China's intelligent computing power scale is expected to reach 52.3%.
"There is still a gap between our country and the United States in terms of intelligent computing capabilities, and the chip problem has also affected the intelligent computing center." An expert in the field of artificial intelligence told Yicai. He explained that although my country currently has an advantage in the scale of computing power , but there are still some gaps in how to make better use of computing power and how to improve the efficiency of computing power.
"Based on the advantages of chips, they are also constantly expanding the scale of computing power, especially the scale of intelligent computing power."
With the introduction of national policy plans such as the "East Counting West" project and new infrastructure, my country's intelligent computing center has set off a construction boom.
At present, more than 30 cities in my country are building or proposing to build smart computing centers. The overall layout is mainly in the eastern region and gradually expanding to the central and western regions.
According to the "Guidelines for the Innovation and Development of Intelligent Computing Centers" recently released by the National Information Center and Inspur Information, during the "14th Five-Year Plan" period, when the intelligent computing center achieves 80% application level, the city's investment in the intelligent computing center can drive artificial intelligence. The core industry increased by 2.9 to 3.4 times, driving related industries to increase by 36 to 42 times.
Developed digital economies such as Beijing, Shanghai, and Guangzhou have also issued policies to promote the construction of intelligent computing centers.
For example, Beijing proposed to "build a batch of computing data centers and artificial intelligence computing power centers, and cultivate them into artificial intelligence computing power hubs by 2023"; Computing power center, promoting the construction of public computing power service platform”, etc.
Just on February 13, Beijing released the "White Paper on the Development of Beijing's Artificial Intelligence Industry in 2022", proposing that Beijing will support leading companies to create large models that benchmark against ChatGPT, and focus on building an open source framework and an application ecology of general large models.
Strengthen the layout of artificial intelligence computing power infrastructure and accelerate the supply of basic data for artificial intelligence.
He Renlong suggested that my country's next focus on the construction of intelligent computing power can speed up the strengthening of computing power as a whole, intelligent scheduling, and strengthen the construction of independent key technologies, especially the research and development of high-end chips, computing systems, software tools and other fields.