Zhang Yunquan, a researcher at the Institute of Computing Technology under the Chinese Academy of Sciences (CAS), could help China break through the U.S.-led stranglehold on the mainland’s AI industry by leveraging the supercomputing technologies it has developed over the past decade, state-run tabloid Global Times reported on Monday. He said at a recent conference that supercomputing systems designed to train large language models (LLMs), the technology behind generative AI services like ChatGPT, are essential to replace power-hungry data center computing clusters that typically use 10,000 to 100,000 graphics processing units (GPUs), the report said. China’s attempt to establish a practical advanced computing platform for training LLMs and developing AI applications illustrates the urgent need for mainland China to become technologically self-sufficient at a time when China’s AI progress remains hampered by limited GPU options amid continuing U.S.-led restrictions. Sanctions have prevented GPU giant Nvidia from supplying cutting-edge chips to the country. Companies are increasingly using data centers – secure, temperature-controlled facilities with large capacity servers and data storage systems – to host or manage the computing infrastructure for artificial intelligence projects. Photo: Shutterstock
“I believe that [building] “LLM cannot be achieved by simply adding more chips,” Chen Runsheng, an academician at the Chinese Academy of Sciences, said at the conference, according to the Global Times. “LLM has to learn to reduce energy consumption while increasing efficiency, just like the human brain.”
According to the report, Chen called on China to engage in fundamental research that combines LLM’s intelligent computing with high-performance computing (HPC) technologies to achieve a breakthrough in computing power. HPC refers to the ability to process data and perform complex calculations at high speeds, which is achieved by supercomputers with thousands of computing nodes working in concert to complete a task.
Engineers work at the Wuhan Supercomputer Center in central Hubei province on May 24, 2023. Photo: AFP
According to Chen, a series of law master’s programs developed in mainland China are based on models and algorithms developed in the United States, without fully considering the basic theory. “If we can make progress in the basic theory, we can achieve breakthrough and real innovation,” Chen said.
