A New Benchmark for Domestic AI Computing Power

At a recent high-profile AI technology conference, Baidu Intelligent Cloud unveiled significant progress in its self-developed computing infrastructure. The announcements highlighted that its indigenously developed Kunlun AI chip has entered a new phase of large-scale application.

Efficient Deployment of 10,000-Card Clusters

Official reports confirm that the Kunlun Chip P800 series has not only passed rigorous large-scale deployment validation but has also successfully delivered multiple super-sized 10,000-card computing clusters since the beginning of 2025. This achievement marks a solid step forward in the chip's industrial adoption.

Powering the Core Training of ERNIE 5.1

A particularly noteworthy accomplishment is the successful completion of the critical training phase for the next-generation ERNIE 5.1 large model on computing clusters built entirely with Kunlun Chips. The training process demonstrated remarkable stability and efficiency.

  • High Training Efficiency: The clusters maintained an impressive effective training rate of 97% throughout the prolonged, high-intensity model training.
  • Superior Scalability: In ultra-large-scale clusters with over 10,000 cards, linear scalability exceeded 85%, with minimal performance degradation.
  • Comprehensive Capability: The infrastructure fully meets the stringent demands of cutting-edge large model training regarding computational precision, operator stability, framework compatibility, and long-duration reliability.

Tianchi 256-Card Super Node Nearing Market Launch

Alongside the training cluster milestones, Baidu shared updates on another key product. The Tianchi 256-card super computing node, powered by Kunlun Chip technology, successfully completed its initial power-on test last month and is scheduled for commercial release in June. This highly integrated super node is designed to deliver more powerful and accessible computing solutions for AI research and development.

These advancements collectively underscore Baidu's commitment to advancing AI chip independence and building a domestic AI computing foundation, injecting fresh momentum into the development of the large model industry.