New AI Training-Inference Integrated Machine based on CPUs were displayed in WAIC

The 7th World Artificial Intelligence Conference (WAIC) kicked off in Shanghai on July 26. This year's conference set a record with an exhibition area exceeding 70,000 square meters, attracting over 800 participating companies and showcasing more than 3,000 cutting-edge products. Highlights included 40+ large AI models, 60+ intelligent robots, and 80+ globally or nationally debuted innovations, making it the largest WAIC to date.

As a leading national team in chip development, Phytium participated in the exhibition with China Electronics Corporation and collaborated with partners such as China GreatWall, DeepGlint, and Songke Intelligent. In the Core Technology Pavilion (Hall H1), the company showcased its cloud-edge-terminal integrated intelligent computing solutions, drawing strong interest from industry clients and ecosystem partners. The debut of the AI training-inference integrated machine based on the Phytium Tengyun S5000C-E and the AI PC powered by the Phytium Tengrui D3000M generated widespread attention.

Phytium has been deeply committed to advancing domestic CPU computing power while embracing the AI revolution, promoting the integration of indigenous chips and AI technologies. Last July, the company partnered with ecosystem allies to launch a fully domesticated intelligent computing center solution capable of supporting large AI models with up to trillions of parameters, offering end-to-end AI solutions for diverse industries.

To date, Phytium's computing platforms have achieved deep compatibility with mainstream open-source large models such as DeepSeek, Qwen (Tongyi Qianwen), and ERNIE Bot, vertical industry models in government, finance, and transportation, as well as dozens of domestic and international AI acceleration cards. Covering cloud, edge, and terminal scenarios, these platforms accelerate industry-wide intelligent upgrades through comprehensive development toolchains, delivering secure and reliable AI computing services to thousands of sectors.

The 4U 8-card AI training-inference integrated machine built on the Phytium Tengyun S5000C-E supports flexible integration of 8–10 domestic or international AI acceleration cards, delivering over 2.5 PFLOPS of computing power—a leading performance level in China. While ensuring autonomy and control, it meets diverse performance demands. Fully compatible with DeepSeek's full and distilled series of large models, the machine covers data center and edge real-time inference scenarios, enabling rapid deployment for government and enterprise users to drive digital transformation.

The AI PC based on the Phytium Tengrui D3000M achieves a maximum clock speed of 2.9GHz, paired with high-frequency LPDDR5 memory and high-speed SSDs. Compared to the previous generation, its performance has doubled, with battery life exceeding 8 hours. Integrated with DeepGlint's AI software, it enables seamless coding, document drafting, and AI-generated PPTs, earning it the title of information technology innovation (ITI) office powerhouse. Leveraging the D3000, D3000M, and S5000C-16 chip series alongside domestic AI acceleration cards, Phytium has developed diverse product forms—desktops, notebooks, and workstations, covering high, medium, and low different configuration requirements, and can meet diversified application scenes such as scientific research, office work, and production.

On the edge computing side, Phytium has established a mature technical ecosystem. By combining its embedded and desktop CPUs with domestic AI acceleration cards in a heterogeneous architecture, the company provides robust support for critical infrastructure industries. For example, bank intelligent teller machines (ITMs) developed with Phytium CPUs and domestic AI cards enable precise facial and iris recognition with identity verification, enhancing transaction security and efficiency. Power grid inspection edge servers based on the same architecture support synchronized monitoring and tagging of 80 video streams, ensuring comprehensive and accurate grid maintenance.

At present, Phytium's cloud-edge-terminal integrated intelligent computing solutions have been deployed across government, finance, power, energy, transportation, manufacturing, and education sectors. A notable example is the first-ever edge-terminal-cloud integrated national AI training and inference platform at Tianjin University, built on Phytium's intelligent computing foundation. This platform meets dual demands for large-scale model cloud training and edge inference deployment, achieving leading technical indicators in the education sector and setting a replicable benchmark for fully autonomous AI teaching and training.

Amid explosive growth in AI computing demand, Phytium is accelerating the development of intelligent computing processors to build a CPU + XPU synergistic architecture, fostering next-generation productivity in high-end chips. The company's first intelligent computing chip is entering its final development phase, poised to deliver diversified computing solutions for major national projects and industry clients.