The 2023 WAIC conference directly hits: the domestic AI large model accelerates the landing computing power chip companies compete to "show their muscles"

Reporter: Xu Yiming Trainee reporter: Sun Wenqing

On July 6, the 2023 World Artificial Intelligence Conference (WAIC) with the theme of "Intelligently Connected World, Creating the Future" kicked off at the Shanghai World Expo Center. On the first day of the exhibition, one of the exhibitors laughed and said, "I'm embarrassed to come to the exhibition if I don't get a little bit involved with the big model."

Since the beginning of this year, AI large-scale models have set off a round of development boom around the world, and are now in a critical period of rapid application landing. The "Securities Daily" reporter learned from the on-site exhibitors that the importance of computing power chip companies has accelerated in the process of implementing large-scale AI models to actual scenarios.

At this exhibition, many exhibitors such as Denglin Technology, Suiyuan Technology, Hanbo Semiconductor, and Mu Xi either launched new products or brought the latest results of their products.

Computing power chip companies are vying to "show their muscles"

Regardless of the number of exhibitors or the exhibition area, the 2023 World Artificial Intelligence Conference is the largest ever. The 50,000-square-meter World Expo main exhibition hall has attracted more than 400 exhibitors, and more than 30 new products have been exhibited for the first time. It is worth mentioning that the large-scale model and the chips and intelligent computing companies that provide computing power support for it are the "bright spots" among the exhibitors.

At the exhibition site, domestic chips such as Hanbo Semiconductor, Suiyuan Technology, and Denglin Technology all used their own products to demonstrate and run large language models, AI painting, Wensheng PPT, etc.

High-end GPU chip company Hanbo Semiconductor has participated in the World Artificial Intelligence Conference for three consecutive years. This time, Hanbo Semiconductor released 6 new products including SG100 full-featured GPU chip, LLM large-scale model AI accelerator card and high-performance generative AI accelerator card in one breath, providing support for AI large-scale model, graphics rendering and high-quality content production Provide a complete solution.

According to the introduction of the company's staff to the "Securities Daily" reporter, the AI large model and metaverse need the support of GPU's powerful AI parallel computing power and graphics rendering power, which has opened up a broad market prospect for cloud-based large chip companies. In the future, GPU chips with parallel computing capabilities will serve as the "brain" of large-scale model calculations, which will provide continuous computing support for large-scale model generation and learning.

Also as a chip design company, Suiyuan Technology released the Wenshengtu MaaS platform service product at this exhibition - Suiyuan Yaotu, which can provide users with fine-tuning and MaaS services for large models and improve the online speed and efficiency of large model applications . Denglin Technology and Tianshu Zhixin were "one step ahead" and directly brought actual cases of application implementation.

The reporter noticed at the technology booth that the company has launched four exhibition areas: large model, intelligent computing center, innovative application and innovative hardware. In the large model display area, Denglin Technology specially showcased the latest generation of innovative general-purpose GPU products-GoldwasserII series and an interactive interface based on open source large language models.

At this exhibition, as the first mass-produced general-purpose GPU company in China, Tianshu Zhixin announced to the outside world that the company's Tiangai 100 chip will continue to complete large-scale model training with larger parameters after completing the training of large-scale model with tens of billions of parameters.

In addition to general-purpose computing power chips, the application of AI models based on high-performance chips is also the "highlight" in the future of the industry. For a long time, Winner Technology has provided standardized or customized data support services for shopping malls, retail brands and other users in the physical business field, as well as government affairs at all levels, through four channels: AI sensors, manual collection, Internet and AI model production. In the next step, the company will promote the digital transformation of more enterprises and government departments based on the existing data base.

** "Per capita computing power" has a bright market prospect**

The popularity of the exhibition is only a corner of the rapid development of domestic large-scale models. In recent years, Beijing, Shanghai, Shenzhen and many other places have introduced relevant support measures to carry out research on large-scale model innovation algorithms and key technologies, strengthen the layout of artificial intelligence computing power infrastructure, and accelerate the supply of basic data for artificial intelligence, thereby building an innovation highland for artificial intelligence.

According to the calculations of the China Academy of Information and Communications Technology, the scale of my country's core artificial intelligence industry will reach 508 billion yuan in 2022, a year-on-year increase of 18%. In the short term, AI large models will undoubtedly become the "new engine" for the development of the core industry of artificial intelligence.

At the World Artificial Intelligence Conference, many computing power chip exhibitors are more optimistic about the localization opportunities in the development of AI large models.

"The development of large-scale model applications directly brings Nvidia into the trillion-dollar market value club. For our domestic manufacturers, there is actually a relatively good development prospect." The above-mentioned exhibitors told the "Securities Daily" reporter that in the future, computing power In the center or other segments, the application of artificial intelligence is bound to be a trend. At least in the next 5-7 years, related companies will still maintain a relatively high-speed growth.

However, different from the development route of overseas large-scale models, the relevant person in charge of the marketing department of Denglin Technology said in an interview with the "Securities Daily" reporter, "We believe that the development of large-scale models in China in the future will be different from overseas. Overseas is mainly for C end, such as Amazon, Google, OpenAI, etc., then there will be many companies in China that mainly focus on the B end, such as applications in medical care and state-owned enterprises. Therefore, the performance, functions, and cost performance of the overall solution are extremely important.”

Faced with the ever-increasing demand for underlying computing power for AI large models, Gai Lujiang, chairman and CEO of Tianshu Zhixin, told the "Securities Daily" reporter that you can now see many companies engaged in large-scale model development, and few need tens of billions of parameters. Many require hundreds of billions of parameters, so the demand for computing power in the future must be more and more. In the words of peers, the future should be based on per capita computing power, that is, how much computing power is consumed every day, just like calculating how much electricity is consumed per person per day now.

A number of brokerages stated in recent research reports that large-scale models have created demand for computing power, and the demand for AI industries continues to explode. The domestic large-scale self-developed AI large-scale model has entered the stage of "letting a hundred flowers bloom", and the emergence of large-scale models has started an "arms race" in computing power. At the same time, the parameters of the large model show an exponential scale, further detonating the demand for massive computing power. As domestic and foreign technology companies continue to invest in the field of AI large-scale models, the chip industry chain related to the AI industry is expected to usher in rapid development.

(Edited by Shangguan Monroe)