e-Tube 技术——突破互连壁垒加速 AI 集群规模化.pdf

《e-Tube 技术——突破互连壁垒加速 AI 集群规模化.pdf》由会员分享，可在线阅读，更多相关《e-Tube 技术——突破互连壁垒加速 AI 集群规模化.pdf（17页珍藏版）》请在三个皮匠报告上搜索。

1、e e-Tube Technology:Tube Technology:Breaking Interconnect Barriers to Breaking Interconnect Barriers to Accelerate AI Cluster Scale UpAccelerate AI Cluster Scale UpPoint2 Technology Inc.e-Tube Technology:Breaking Interconnect Barriers to Accelerate AI Cluster Scale UpDavid KuoVP of Product Marketing

2、Point2 TechnologyNETWORKINGEnergy Efficiency Critical to Sustainable AI GrowthAI-Focused Interconnect is Vital*Commscope-Cabling Considerations of AI data centersNetwork Latency=Compute Efficiency=Energy EfficiencyDramatically ReduceInterconnect Power Consumption1Higher Bandwidth Density to Unblock

3、GPU-To-GPU Communication2Near-Zero Latency to Minimize GPU-to-GPU Communication Delay370%30%LLM Training Cost*ComputeNetwork LatencyAI Cluster Scale-Up Interconnect Targets MetricTargetEnergy Efficiency 5 pJ/bCable Reach 10mCost$0.10/Gbps$0.05/GbpsBeach Front Density 1 Tb/mmLatencyNear zeroThe Coppe

4、r CliffToo thickToo shortNot scalable5Copper and Optical Barriers Limiting Short-Reach,Terabit-level AI Compute Scale UpThe Optical PenaltiesToo ExpensiveToo power hungryReliability Issues LatencyIntroducing e-TubeBreaking the Barriers of Copper and Optical CablingPower3X LowerLatency1000X LowerCost

5、3X LowerCable Length10X Coppervs.OpticalCablingRF Transmitter and Receiver ICs for Data Transmission over Plastic Dielectric Waveguidee-Tube flat frequency response and fixed bandwidth across the spectrum Scale in speed using the same e-Tube core construction for 224G/448G/beyondHow e-Tube WorksElec

6、trical Domain(Copper)Electrical Domain(Copper)RF Domain(eRF Domain(e-Tube over Plastic Waveguide)Tube over Plastic Waveguide)Cable GenerationSERDES Data RateModulationDifferential Pairs(TX&RX)Dual Carrier FrequenciesChannel BandwidthModulation RF Data Ratee-Tube Cores(TX&RX)1.6T224GPAM416(8 Tx,8 Rx)