Shulou(Shulou.com)12/24 Report--

On November 24, the project of China Unicom Yangtze River Delta (Wuhu) Intelligence Center, with an investment of 6 billion yuan, began.

On November 22, the Art (Wuxi) Intelligence Center for Automotive Research and Development was officially opened.

On November 21, China Mobile Intelligence Center (Wuhan), which planned the first phase of 1000PFLOPS computing capability, landed.

On November 20, the first domestic Intelligent Computing Center project was officially signed in Gui'an.

In addition, such as Ordos, Nanchang and other places, there are intelligent calculation-related land planning. News of related projects has been reported almost every day in the past week. This is almost the norm in the past year, and the construction of the Intelligence Center has entered an explosive period.

Behind the popularity of smart computing is the rise of the AI model and its new demand for computing power.

How horrible is the calculation power consumption of the big model?

With the rapid development of artificial intelligence technology, AI large-scale model has become the main driving force for the growth of computing power in the world. These models not only make a breakthrough in theoretical research, but also play a more and more important role in practical application. The growing popularity of AI models is driving changes in computing infrastructure around the world and in China.

We can get a glimpse of how much computing power is required by the large model of AI from the training power consumption of ChatGPT. OpenAI's report shows that training a 174.6 billion-parameter GPT-3 model requires about 3640PFlops-day. Although details of ChatGPT 4 have not been released, ARK Invest estimates that the number of GPT-4 parameters is as high as 1.5 trillion, and computing power requirements can be as high as 31271 PFlops-day.

This is only the training demand of ChatGPT-after ChatGPT has attracted wide attention, countries all over the world are entering the field of large models, and all kinds of large model products emerge in endlessly. In less than a year, no less than 200 large model products have been released in China alone.

You can imagine how scary the computing power required by these large models will be.

Challenges to the Computing Infrastructure

The surge in demand for numeracy poses challenges to the computing infrastructure in two directions.

On the one hand is the challenge of computing power production. With the advent of the limit of Moore's Law at the level of physical law, relying on the computing power of a single chip can no longer meet the needs of super-large-scale computing power, so more chips are needed to work together, so the scale of the data center is getting larger and larger. The construction of clustered computing centers has become the norm.

On the one hand, there is the challenge of the environment needed to produce computing power. In the case of the super-large-scale development of the computing center, the adequacy and stability of the energy supply of the computing center, and how to effectively distribute the heat caused by the huge energy consumption has also become a realistic topic that needs to be faced.

And these are only the computing power needed in the training of AI large models. With the landing and popularization of large model products, the demand for inference-based computing power will enter a period of growth, and its increment will exceed the training demand. In the long run, the total demand for numeracy will be on a par with or even exceed the training demand.

Therefore, the new demand for computing power brought about by the AI model is not only a technical challenge, but also an important opportunity for infrastructure development. Especially in the Chinese market, the rapid growth of the data center industry indicates great potential and opportunities. At the 18th China IDC Industry Annual ceremony to be held from December 12 to 14, a number of venues and topics will be set up, and experts and enterprises from upstream and downstream of the smart computing industry, such as Kuaishou, Century Interconnection, Superfusion, iFLYTEK, Smart Spectrum AI, and BYD, will be invited to discuss how to face the computational challenges.

-

The 18th China IDC Industry Annual Gala (IDCC2023) will be held in Beijing National Convention Center on December 12-14, 2023. The theme of the conference is "Digital Computing Evolution", focusing on hot topics in the computing industry and bringing together elites in the industry to discuss the development direction and future trend of the digital economy. Welcome to sign up and participate on the spot!

Scan the code and sign up immediately

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.