当前位置:首页 > 报告详情

20260219_A-201_Lewis.PDF

上传人: 彩旗 编号:1158848 2026-03-02 14页 1.52MB

1、Telemetry as the Backbone of AI-Ready Infrastructure12Confidential|2025 Insyde Software2 2Company AI Compute Demands More Than Compute Explosive growth of AI workloads in cloud&edge.The global AI data centers market size is expected to reach nearly USD 165.73 billion by 2034 increasing from USD 17.5

2、4 billion in 2025 and accelerating at a solid CAGR of 28.34%from 2025 to 2034.*Along with that growth,there is increased need to manage the remote nodes and monitor the increased flow of being generated.Monitoring of remote nodes requires telemetry infrastructure to gather,summarize and react to eve

3、r-changing conditions within the data centers.2 *AI Data Centers Market Overview,Precedence Research,9/10/2025Arm Server Manageability ArchitectureBMC(Baseboard Management Controller)System InterfaceRedfish and IPMI EnginesSoC Side-band InterfacePlatform Elements InterfaceIO Device Side-band Interfa

4、ceArm based Server SoCHost OS/FirmwareAMC/SatMCPlatform ElementsIO DevicesBMC OOB management servicesAdminHost/SoC in-band interface IPMI and Redfish HI Sensors,FAN,Power etc.,Networking,Storage,GPUs and DPUs etc.Platform elements interfacesBMC in-band management servicesIPMIIPMIAdminHost AMC interf

5、aceSoC side-band interface(MCTP/PLDM/SPDM)BMC IO device interface(MCTP/PLDM/SPDM)Arm Server Base Manageability Requirements(SBMR)specification defines the minimum architecture for Arm server managementArchitecture includes Auxiliary Management Controller(AMC)(or Satellite MC-SatMC)embedded on the So

6、CSBMR also defines interfaces between the sub-systems4Confidential|2025 Insyde Software4 4Company Telemetry:The Nervous System of Modern DatacentersBMC(Baseboard Management Controller)System InterfaceRedfish and IPMI EnginesSoC Side-band InterfacePlatform Elements InterfaceIO Device Side-band Interf

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. **AI算力需求激增**:全球AI数据中心市场规模预计从2025年175.4亿美元增至2034年1657.3亿美元,年复合增长率28.34%。 2. **遥测是核心基础设施**:需通过遥测系统收集、汇总远程节点数据,确保AI工作负载的可靠性与早期故障检测。 3. **挑战与瓶颈**:SoC核心数增长(如256核导致768个传感器)引发数据过载,需优化网络流量与响应速度。 4. **合作与解决方案**:Arm与Insyde合作,基于开源模型(如OpenBMC、FVP)实现从概念到生产的遥测方案,支持硬件仿真与软件开发。 5. **AI与遥测闭环**:结合AI分析遥测数据,实现预测性维护与数据中心自主优化,推动AI基础设施向自管理演进。
**AI基建瓶颈?** **遥测如何赋能AI?** **自管系统可行吗?**
客服
商务合作
小程序
服务号
折叠