Since the beginning of this year, with the rapid recognition of AI large models, the global intelligent industry has ushered in a new peak of development. Major technology manufacturers around the world have laid out large-scale models one after another, and there has been a grand occasion of "Hundred Models War" in China.
With the rapid rise of large models, a problem that has been lingering in the AI industry for many years has been further exposed: the demand for special computing power for AI tasks continues to increase, and the supply continues to be in short supply.
All of a sudden, the phenomenon of "computing panic" and "difficult to find a card" generally appeared in countries all over the world, and there was even a strange phenomenon that European and American technology companies began to use GPU resources for financing.
These phenomena clearly tell us that in the age of intelligence, computing power is productivity, and computing power is capital. Facing the era of intelligence, how to solve the challenge of computing power difficulty and shortage? How to make a long-term and healthy development of intelligent computing power technology? How to make intelligent computing power a boost for technological self-reliance and self-improvement?
These questions require a definite answer.
On August 18, as China's first provincial-ministerial joint meeting with the theme of data center computing power empowerment, the 2023 China Computing Power Conference opened in Yinchuan, Ningxia. Development" as the theme, showing in detail the development prospects and latest achievements of China's computing power industry.
Among them, H3C Group, a subsidiary of Tsinghua Unigroup, brought a new generation of smart computing products, technologies and solutions to the scene. At the meeting, H3C and China Academy of Information and Communications Technology jointly released the "2023 White Paper on the Development of Intelligent Computing Power".
Facing the turbulent era of intelligence, in the China Computing Power Conference, you can find and confirm the "Chinese answer" about smart computing power.
Definition and Trend of Intelligent Computing Power
Facing the turbulent AI wave, everyone may have heard of the rising GPU, but few people may be able to say in detail, what exactly is intelligent computing power?
From the perspective of basic technology, intelligent computing power, also known as artificial intelligence computing power, is a computer system capability for artificial intelligence applications that provides artificial intelligence algorithm model training and model operation services.
As the importance of AI technology continues to increase, the value of intelligent computing power as infrastructure is also increasing. It can be said that intelligent computing power is the foundation of intelligent applications and intelligent industries, and a prerequisite for developing an intelligent economy and building an intelligent society.
No matter from various aspects, intelligent computing power is in a period of rapid explosion. By the end of 2022, the total scale of global computing power will reach 650 EFLOPS, of which the scale of intelligent computing power will be 142 EFLOPS, accounting for an astonishing 21.9%, an increase of 25.7% compared with last year. It can be said that in the entire computing industry landscape, intelligent computing power is currently the fastest growing, the highest demanded, and the most concerned.
As for the development of China's intelligent computing power, it can be seen that by the end of 2022, the total scale of China's computing power will be 180 EFLOPS, and the scale of intelligent computing power will be 41 EFLOPS. In other words, the scale of intelligent computing power accounted for 22.7%, an increase of 41.4% compared with last year. This growth rate far exceeds the average growth rate of global intelligent computing power. In other words, China's smart computing power industry is currently in a stage of development with large demand, fast growth, and high value. It can be seen that semiconductor and computing companies have aimed at the wave of intelligent computing power and launched targeted products and solutions. For example, H3C has launched an AI server for large-scale model training, a 51.2T, 800G CPO silicon photonic data center switch, and an Aofei computing power platform that supports large computing power scheduling.
After understanding the schedule and development speed of intelligent computing power, the next question is, what is the ultimate application value of intelligent computing power?
Judging from the current situation, it seems that the most popular technological outlets such as Metaverse, autonomous driving, AIGC, and digital twins cannot do without the support of intelligent computing power. For example, AIGC is currently a very hot field, setting off a wave of investment and construction around the world. Behind AIGC is the organic integration of core elements such as computing power, data, and algorithms. The larger the model used, the higher the computing power requirements. Judging from the current trend, AIGC's demand for intelligent computing power will be stronger in the future, and the value of intelligent computing power-related industries will be reshaped.
At the 2023 China Computing Power Conference, it can be seen that all walks of life are paying attention to the solutions and supply of smart computing power, looking forward to obtaining more powerful support for smart computing power. This is because intelligent computing power is the source of intelligent technology and applications, and the base of intelligent innovation.
Next, we can go deeper and explore more possibilities on the road to the development of intelligent computing power.
Challenges and Opportunities Faced by Intelligent Computing Power
As far as any technology industry is concerned, it must continue to complete the development in the cycle of "encountering challenges-solving problems-industrial upgrading", and intelligent computing power is no exception. It can be seen that in the case of a huge demand for intelligent computing power emerging in the global technology industry at the same time, there has been a significant shortage of supply in this field, which has caused phenomena such as "computing power panic" and "computing power difficulty". At the same time, with the continuous development of AI technology, issues such as energy consumption, security, and ecological cooperation are gradually exposed. But only when there are challenges can there be development goals and opportunities for relevant companies and practitioners.
The "2023 White Paper on the Development of Intelligent Computing Power" pointed out that the development of intelligent computing power has five major challenges. At the same time, the "White Paper" also gives a detailed solution to each challenge, helping the industry to consolidate consensus and capture the development window.
For example, the demand for computing power is the core of intelligent computing power, and it is also the most important challenge. The sharp increase in the demand for intelligent computing power is a common problem faced by the global AI field. In order to cope with the shortage of computing power including intelligent computing power and promote the intensive development of socialized computing power, China creatively proposed and implemented the East-West-West Count engineering. At the same time, we must also see that the linkage challenges between data centers have caused the efficiency of computing resource allocation to still need to be improved. Solving these problems constitutes a major opportunity for the development of intelligent computing power and the continuous advancement of east and west calculations.
To this end, the "White Paper" proposes a solution to upgrade the supply of computing power resources by building intelligent computing centers and cloud computing centers, and at the same time dispatch idle computing power around data sources through the network. Combined with solutions such as improving the efficiency of computing power usage, developing a new network architecture, and establishing a unified computing power scheduling platform, it can maximize the value of east-west computing and meet the huge demand for intelligent computing power from all walks of life in China.
In addition to the demand for computing power, the "White Paper" also pointed out that the challenges of energy consumption continue to increase; the complexity of algorithms continues to increase; the data privacy and security issues faced by AI models and AI computing; the ecological cooperation problems related to intelligent computing power industries, etc. A series of challenges, and specific solutions to these challenges are given. It can be said that at this stage, intelligent computing power is still in the initial and golden period of rapid development. Only by identifying the challenges and working efficiently can we maximize the dividends in the development of intelligent computing power.
Facing the future, China's answer
After summarizing the definition and examining the challenges, the next most critical question is: How should China's technology industry break through the development of intelligent computing power? How to give answers that conform to the trends and trends of the times.
To this end, the "White Paper" provides an in-depth perspective on the future development of intelligent computing power, and provides development suggestions for various related industries. It pointed out that the development of intelligent computing power will have five basic trends in the future:
1. The penetration of artificial intelligence is accelerating, and diverse scenarios have spawned diversified computing power requirements.
Facing the future, AI technology will not exist alone, but deeply penetrate into scenarios and applications. Therefore, heterogeneous and diverse computing demands will continue to increase, and intelligent computing with cloud-edge-device collaboration will become the mainstream. Driven by these factors, we will enter a new era of diverse computing.
2. Driven by policies, the low-carbon development of intelligent computing power has become a rigid requirement.
The development of intelligent computing power happens to be at the same frequency as the implementation of the dual carbon goals. Therefore, in the future, under the guidance of policies, a more carpeted computing power acquisition method will be adopted to meet the needs of AI. Low-carbon technology will also become a core competitiveness of the computing and AI industries.
3. Edge intelligence emerges as the times require, and edge computing and artificial intelligence are integrated and developed.
Relatively speaking, obtaining intelligent computing power from the cloud will greatly depend on the network environment, and localized intelligent computing power faces high hardware costs. Therefore, in the future, it is very likely that the way of acquiring intelligent computing power will be migrated to the edge side, so AI technology will develop simultaneously with the edge computing industry.
4. The construction of the intelligent computing center is accelerated to meet the demand for high-quality computing power.
Under the trend of east-west computing and new-generation data centers, intelligent computing centers that provide heterogeneous computing resources will become a new outlet. The Intelligent Computing Center will provide an exponential increase in intelligent computing power to achieve performance upgrades and energy consumption reduction in AI scenarios.
5. The scale of the model continues to expand, and the massive and diversified data urgently requires huge quantitative computing power.
Since the rise of deep learning, a clearly visible rule is that the scale of the model is getting larger and larger, and the effect of the emergence of intelligence brought about by the scale of the model is becoming clearer. There is reason to believe that a huge number of models with very large parameters will appear next. The current large model may be just the beginning, and intelligent computing power must be prepared for this.
Under the pressure of these five major trends, the intelligent computing power industry will usher in a deterministic future development. First of all, the demand for computing power will continue to explode, and the momentum of industrial development and the challenges of the supply chain will continue to increase; secondly, the localization process of intelligent computing power will inevitably accelerate, and the full-stack independent intelligent computing power will become the focus and focus of the next industrial development. Strategic pillar; In addition, intelligent computing power will be integrated into thousands of industries to create value, which also provides impetus for intelligent computing power solutions and industrial ecological development in various fields.
For this reason, the "White Paper" puts forward suggestions for the development of intelligent computing power in terms of industry, technology, and standards.
For example, in terms of industry, it is recommended to formulate plans from the national strategic level, and take multiple measures to promote the healthy and orderly development of intelligent computing power. These include building a unified intelligent computing power service center and incubation platform, strengthening financial support for related industries, and promoting the green development of intelligent computing power.
In terms of technology, we should increase the research and development of intelligent computing power, especially the innovation of key technologies. In terms of standards, accelerate the construction of open standards, and transform diversified computing power into schedulable computing power resources.
According to IDC's prediction, the scale of China's intelligent computing power will continue to grow rapidly. It is estimated that by 2026, the scale of China's intelligent computing power will reach 1271.4 EFLOPS, and the compound growth rate will reach 52.3% in the next five years. It can be said that intelligent computing power is ushering in its golden period of development in China, and related various technical fields and industrial ecological roles will usher in historic development opportunities.
The vast river of intelligent computing power requires us to understand, face, and explore. In the end, we will sail on it and give China's answer about intelligent computing power and the era of intelligence.
