当前位置:首页 > 报告详情

European 1MW rack - HDINI.pdf

上传人: S** 编号:1240928 2026-05-16 13页 1.86MB

1、European 1MW rack-HDINILuis Castrillo Senior Technologist at CastrolJon Summers Scientific Leader in Data Centers at RISECOOLING ENVIRONMENTS&LIQUID COOLINGTrends in power density,influence of AI models and workloadsConnect AI demands to 6G edge infrastructureInvesting in Cloud,Edge and the Internet

2、 of Thingshttps:/digital-strategy.ec.europa.eu/en/policies/iot-investingChallenges that need to be solved for this infrastructure to be deployedPower delivery with minimal losses for 1MW in a rack volume.Distribution of liquid cooling at high density.Emulation of next generation dense compute.Minima

3、l losses in the thermal chain from the chip to the point ofheat rejection/reuse.Stressing thermal energy recovery.The characters of this storyLets focus on the cooling-how we will solve itOn chipImmersionIn-house developed Chip Load Emulator by RISEHybrid liquid cooling(1P Immersion+1P DTC)On chipDT

4、C HEX1Imm HEXDTC HEX2405060Optimising for heat recovery exampleTesting rig to be ready soonThermal stack optimisation from TIM2 to heat rejection modelling is essentialChip(Silicon)TIM1IHSTIM2Heat sink/Cold plateCDUExternal Heat rejection/reuseunitTambient/rejectionTjunctionTcaseTcold plateTTCS_supp

5、lyTFWS_supplyTsiTTIM1Potential challenges and roadmapDTC only vs DTC+Immersion2025(H2)2028(H2)202720262028Performance vs efficiencyImprovement and cost of heat reuseActual power deliveryA single rack dissipating 1MW optimised for high-density compute and energy efficiencyPowerCoolingPower distributi

6、onServer coolingCooling distributionHeat reuse design1MWEuropean commitment to innovate and not only regulateWhen considering future 1MW rack deployments consider the innovation fromthe H-DINI project.Timeline to a demonstration is envisage

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. **项目目标**:欧洲H-DINI项目聚焦1MW高密度机架创新,解决功率传输、液冷分布及热能回收等挑战,目标2028年下半年演示。 2. **核心技术**:采用混合液冷(1相浸没+1相直接接触冷却),结合自研芯片负载模拟器,优化从芯片到热排出的热链路。 3. **关键数据**:单机架功耗1MW,需实现功率传输零损耗、液冷高密度分布及热能回收最小化损失。 4. **进展与协作**:依托RISE ICE测试中心,参与OCP相关子项目(如冷板、浸没等),推动6G边缘与AI基础设施融合。
**1MW冷却挑战?** **AI如何驱动冷却?** **热回收能省钱?**
客服
商务合作
小程序
服务号
折叠