当前位置:首页 > 报告详情

面向人工智能基础设施的线性光器件:单通道 112G 和 224G 的高效节能 LPO 和 LRO 解决方案.pdf

上传人: 明**** 编号:1012042 2025-12-21 16页 2.22MB

1、Tiger Ninomiya,Principal TechnologistAccelink USALinear Optics for AI Infrastructure:Power-Efficient LPO and LRO Solutions at 112G and 224G per LaneOCP SPECIAL FOCUS:PHOTONICSTiger Ninomiya,Principal TechnologistAccelink USALinear Optics for AI Infrastructure:Power-Efficient LPO and LRO Solutions at

2、 112G and 224G per LaneOCP SPECIAL FOCUS:PHOTONICSAI Bandwidth&Power ChallengesPluggable Optics Options:DSP,LPO,LROStandardization&Ecosystem ReadinessOutlineAI Cluster Architecture Optics are essential to enable AI scale-out across servers and racks Scale-up applications may also adopt optics as cop

3、per reach limits are challenged Optics are essential to enable AI scale-out across servers and racks Scale-up applications may also adopt optics as copper reach limits are challengedAI/ML Bandwidth&Power DemandsReference:IEA(2025),Global data centre electricity consumption,by equipment,Base Case,202

4、0-2030,IEA,Paris https:/www.iea.org/data-and-statistics/charts/global-data-centre-electricity-consumption-by-equipment-base-case-2020-2030,Licence:CC BY 4.0 AI/ML growth drives the need for higher bandwidth,with 224G/lane becoming critical Greater energy efficiency is essential to manage the rising

5、data center power demand AI/ML growth drives the need for higher bandwidth,with 224G/lane becoming critical Greater energy efficiency is essential to manage the rising data center power demandGlobal data centreelectricity consumption,by equipment,Base Case,2020-2030-10.00 20.00 30.00 40.00 50.00 60.

6、00 70.00 80.00 90.002021202220232024202520262027202820292030100G200G400G800G1.6T3.2TSales of Ethernet transceivers by data rate(total market)Reference:LightCountingPluggable Optics Options for AI InterconnectsTraditional Pluggable Optics(DSP)Linear Pluggable Optics(LPO)Digital Signal Processor(DSP)b

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。
根据文章内容,以下是全文关键点的概括: 1. **AI基础设施光学需求**:AI和机器学习(AI/ML)的增长推动了对更高带宽的需求,224G/lane成为关键,同时需要更高的能源效率来管理数据中心不断增长的电力需求。 2. **光学解决方案**:文章讨论了三种光学解决方案:传统可插拔光学(DSP)、线性可插拔光学(LPO)和线性接收光学(LRO/TRO/RTLR)。 3. **性能与效率**:LPO和LRO在降低功耗和提高效率方面优于DSP,例如LPO模块温度比DSP低约15℃。 4. **标准化与生态系统**:IEEE 802.3dj和LPO MSA等标准推动了112G的部署,并促进了224G的可行性。 5. **技术优势**:DSP提供全面均衡和信号完整性,但功耗高;LPO和LRO在降低功耗和延迟方面表现优异,但需要更紧密的链路工程。 6. **总结**:线性光学和DSP结合,为可扩展、节能的AI基础设施提供了支持。
"AI集群,光学如何助力?" "112G/224G,光学解决方案揭秘!" "数据中心电力需求,光学节能之道!"
客服
商务合作
小程序
服务号
折叠