当前位置:首页 > 报告详情

DC25_SLIDES_Track1_ANovelThermoFlowUniformizerApplied_Son.pdf

上传人: S** 编号:1240857 2026-05-16 38页 10.18MB

1、Information Classification:GeneralWelcome to ConferenceJanuary 2830,2025Santa Clara Convention Center1ExpoJanuary 2930,2025Information Classification:GeneralA Novel ThermoFlow Uniformizer applied Immersion Cooling System(TFU-ICS)for Highly-Dense Multi-GPU-HBM Based Compute Module for AI Supercompute

2、rKeeyoung Son(KAIST)2Information Classification:GeneralSPEAKERS3Keeyoung SonPh.D.,KAISTKeeyoung Son received the B.S.degree in mechanical engineering from the Korea advanced Institute of Science and Technology(KAIST),South Korea,in 2019.and the M.S.and Ph.D.degrees in electrical engineering from the

3、 KAIST,in 2021 and 2025,respectively.He has research experience about six years in the field of signal/power integrity(SI/PI)for high-speed interconnects and thermal effects on SI/PI.His professional experience includes serving as an HDMI consultant at Analog Devices and working as an RFIC EM team i

4、ntern at Qualcomm in San Diego.His current research focuses on the design and analysis of SI/PI in electronic packaging systems.https:/ Classification:General Introduction Proposal of ThermoFlow Uniformizer applied Immersion Cooling System(TFU-ICS)Thermal Evaluation of TFU-ICS for Multi-GPU-HBM base

5、d Highly Dense Compute Module Temperature-dependent Signal Integrity Evaluation of Compute Module with TFU-ICS Conclusion4OutlineInformation Classification:GeneralMulti-GPU-HBM based AI Supercomputer and Thermal Management Issues5Thermal issue from dense design ofCompute moduleRackInformation Classi

6、fication:GeneralNecessity of Powerful and Uniform Cooling System for Multi-GPU-HBM based Compute Module6Information Classification:GeneralHeat Sink Development with Increased Power Consumption7 As power consumption of system has increased,heat sink developed for high cooling performance.For thermal

word格式文档无特别注明外均可编辑修改,预览文件经过压缩,下载原文更清晰!
三个皮匠报告文库所有资源均是客户上传分享,仅供网友学习交流,未经上传用户书面授权,请勿作商用。
1. **研究背景**:针对AI超算中多GPU-HBM高密度计算模块的热管理问题,传统散热(如风冷)无法满足GPT5级模型需求,需先进浸没冷却系统(ICS)。 2. **方案提出**:设计ThermoFlow Uniformizer-ICS(TFU-ICS),通过调节TFU厚度控制冷却液流速,提升散热效率并降低泵功耗。 3. **性能优势**:TFU-ICS实现更均匀的温度分布(标准差σT更低),支持硅中介层集成多GPU-HBM,提高计算密度;温度依赖信号完整性评估显示,GPU-GPU/GPU-Switch链路无需均衡器即可满足眼图规范。 4. **核心数据**:TFU设计基于最大功耗密度(如GPU区域)优化,双面冷却使模块温度均值更低,且泵功耗因TFU尾部减薄而降低。
**TFU如何降温?** **AI散热新方案?** **高密度计算可行吗?**
客服
商务合作
小程序
服务号
折叠