Baidu makes its debut with Kunlun chip super node, single card performance can be improved by 95%
At the 2025 Baidu World Conference, Baidu showcased the Kunlun Chip Supernode for the first time, emphasizing its innovations in "core technology." This supernode has been fully deployed in Baidu Baike 5.0, supporting the industry's largest-scale 1 trillion parameter open-source model to run in just a few minutes. By integrating multiple Kunlun Chip AI acceleration cards, single card performance has increased by 95%, and single-instance inference performance has improved by up to 8 times
The 2025 Baidu World Conference is being held today. At the "Baidu World Exhibition Area" held concurrently with this conference, Baidu's Kunlun Chip Super Node made its public debut, showcasing Baidu's "Core Technology." According to insiders at Baidu, the Kunlun Chip Super Node has been fully deployed in Baidu Baijie 5.0 since April of this year and has achieved large-scale deployment. With this node, the industry's largest 1 trillion parameter open-source model can be run in just a few minutes through a single cloud instance. Additionally, by integrating multiple Kunlun Chip AI acceleration cards into a unified super node architecture, a 95% performance improvement per card has been achieved under the optimization of the DeepSeek V3/R1 PD separation inference architecture, with single-instance inference performance significantly increased by up to 8 times. (Sina)
Risk Warning and Disclaimer
The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investment based on this is at one's own risk
