AI News

Computing power as the foundation, AI enters the laboratory — new trends in AI development seen from the 23rd China Computer Conference

Jiwei.comUpdated Oct 25 19:56

AI summary

The 23rd China Computer Conference opened in Chengdu on October 22, with more than 11,000 registered attendees. At the conference, the "Tianfu·Huiyan" integrated platform for AI-enabled scientific and technological innovation was officially launched, centered on the "Tianfu·Huiyan" 1.0 scientific foundation model with approximately 35 billion activated parameters. Li Bin, Senior Vice President of Sugon, introduced that the Sugon 8000 (Dengfeng) fully domestic 100,000-card AI supercluster, launched in July this year, has supported simulation of more than 300 key scientific engineering projects and pre-training of 10,000-card-level models. In the first three quarters of this year, Sichuan's AI industry revenue reached 152.1 billion yuan, up 21% year on year.

Body · AI translation

Computing power becomes the foundation, AI enters the laboratory — new trends in artificial intelligence development as seen from the 23rd China Computer Congress

On October 22, the 23rd China Computer Congress opened at the Western China International Expo City in Chengdu, with more than 11,000 registered attendees. This is the first time in the more than two decades since the China Computer Congress was founded that it has been held in Sichuan.

Inside and outside the venue, three words were repeatedly mentioned: data, computing power, and laboratories.

"As model parameters grow and agents are deployed at an accelerating pace, computing power systems are moving toward intensification and large-scale operation, requiring integrated, tightly coupled system innovation around computing, storage, interconnection, power supply, and cooling," said Li Bin, senior vice president of Sugon. Through "super node + super cluster" system-level innovation, Token production efficiency can be greatly improved, he said.

Li Bin introduced that the Sugon 8000 (Dengfeng) fully domestic 100,000-card AI super cluster, launched in July this year, has already supported simulation for more than 300 key scientific projects and pre-training of 10,000-card-level models.

"Large model parameter counts are leaping from the hundreds of billions to the trillions, or even ten trillions, and a single card or single machine can no longer support large-scale training and inference," said Sun Jianpeng, an intelligent computing quantitative design expert at Hygon. AI infrastructure construction is shifting from optimizing individual components to complex system engineering coordination; only by coordinating computing, communication, storage, and the scheduling of chips and software can overall system efficiency be improved, he said.

AI applications are also shifting gears. Fan Qiwu, chief technology officer of Sichuan Provincial Natural Resources Digital Technology Co., Ltd., introduced that the mining large model's mineral exploration stage led by the company has been applied in more than ten units, and the mining stage is being trialed at the Hongge South Mine and mines under Sichuan Gold, with all construction expected to be completed by the end of next year to achieve on-site production with unmanned mining truck scheduling.

"In rock and mineral identification, traditional manual identification requires a graduate degree plus 2 to 3 years of experience, processing more than 10 slides per day on average; after training with 50,000 high-quality annotated data samples, AI processes thousands or even tens of thousands of slides per day on average, with accuracy rising from 85% for manual work to 93%, and is expected to create economic value of tens of millions of yuan annually," Fan Qiwu said.

At this congress, AI for Science became a hot topic among scientists.

"For AI to enter the laboratory, the first hurdle is data," said Zhu Xiangdong, a professor at Sichuan University and head of the pilot test base of the National Engineering Research Center for Biomaterials. "Materials science data has long suffered from the problem of 'scattered resources and lack of coordinated standardization.'" At present, focusing on advantageous disciplines such as biomedical materials and functional composite materials, Sichuan has standardized the collection of data across the entire chain of literature, computation, experimentation, production, and service.

During the congress, the "Tianfu · Huiyan" integrated platform for AI-enabled scientific and technological innovation, built by Sichuan Province, was officially launched. With the "Tianfu · Huiyan" 1.0 scientific foundation large model, which has about 35 billion activated parameters, as its core, and supported by domestic computing power, scientific data, and dry and wet intelligent laboratories, the platform will expand into N application scenarios in key fields such as new materials, new drug development, and clean energy, connecting the entire process of prediction, experimentation, verification, and feedback, so that scientific research can move from repeated trial and error to intelligent iteration.

It is understood that Sichuan is advancing "AI+" as the province's "No. 1 Innovation Project." In the first three quarters of this year, it achieved industrial revenue of 152.1 billion yuan, up 21% year on year. The "Sichuan Artificial Intelligence Industry Innovation and Development Report" released at the congress shows that in 2025 Sichuan's AI industry revenue reached 191.97 billion yuan, up 22.9% year on year; the province has 2,806 key AI enterprises, including 2,380 high-tech enterprises and 93 specialized, refined, distinctive, and innovative "little giant" enterprises.

Source

Jiwei.com