Computing network a bid to pool resources

Several provinces join scheme to consolidate digital power amid tech race over AI capability

China’s national computing power network has accepted its first batch of provinces to join in, as Beijing seeks to pool and build up national digital infrastructure despite limited access to advanced chips amid US-imposed trade restrictions.

Data centres in Guangdong and the provinces of Sichuan and Guizhou have been admitted to the China Computing Net (C2NET), according to a report by government-backed newspaper Southern Daily that cited a Monday announcement at an industrial event in Shaoguan, Guangdong.

The C2NET was unveiled last May to better consolidate and allocate computing power. By pulling together regional resources, the network operated with a coordinated computing power of more than 3 exaflops, meaning the ability to perform 3 quintillion calculations per second, the report said.

In comparison, the world’s top supercomputer, the Frontier system in the United States, can achieve a maximal performance of 1.2 exaflops, according to ranking agency TOP500.

Demand for computing power is skyrocketing thanks to the rising popularity of artificial intelligence (AI), and China is trying to ensure sufficient computing power to support the country’s research and technological progress.

“Computing power will become a type of utility like electricity that no matter where the supply is, one can pay to use it when there is demand for computing,” said Gao Wen, director of the Shenzhen-governmentowned Peng Cheng Laboratory, which leads the C2NET project.

Gao added that computing power was also key to developing large language models (LLMs), the technology used to train the AI-powered chatbot ChatGPT.

“The combination of computing power, data and algorithms is the core of LLMs,” Gao said, according to media outlet Caijing.

Historian Chris Miller, the author of Chip War, has stated that “the rivalry between the United States and China may well be determined by computing power”.

Computing power has become a key factor for a country’s technological growth, and the US has been restricting China’s access to advanced chips in recent times.

For example, Nvidia’s data centre chip the A100 was added to the US export control list in August 2022. It is also uncertain whether China will have access to Nvidia’s GH200 super chip, its most powerful AI chip yet, which the manufacturer said on Monday was now in full production.

Chinese companies and research institutes including Baidu and Alibaba Group Holding have rushed to launch domestic alternatives to OpenAI’s ChatGPT. Alibaba owns the Post.

Chinese institutions have so far launched at least 79 LLMs with 1 billion parameters, a measure of the size and complexity of an AI model. In comparison, OpenAI’s GPT-3.5 model has 175 billion parameters.