Haiguang Information officially announces the "Intrinsic Security" technology achievement, racing towards the training of domestically produced trillion-parameter large models

robot
Abstract generation in progress

ME News Report, April 3rd (UTC+8), on April 2nd, at the Haiguang Information 2026 Spring Technology Communication Conference, Haiguang officially announced several new achievements based on the “Endogenous Security” concept, and debuted the annual version of the Haiguang DCU software stack, providing computing power support for the training of multiple domestic trillion-parameter large models. Haiguang Information Vice President Ying Zhiwei stated that the security challenges of computing power in the AI era have surpassed the capabilities of traditional solutions and must seek breakthroughs from the very bottom of the computing architecture. On a computing cluster composed of tens of thousands of Haiguang CPUs and DCUs, the Haiguang team has preliminarily achieved stable operation of models with 100 trillion parameters through multiple technologies such as endogenous security, operator tuning, compiler optimization, and integrated computing, with several tests including MLPerf reaching internationally leading levels. (Source: Ifnar)

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin