专题讨论会 - GPU 管理标准化：Redfish、遥测和固件更新协议.pdf

《专题讨论会 - GPU 管理标准化：Redfish、遥测和固件更新协议.pdf》由会员分享，可在线阅读，更多相关《专题讨论会 - GPU 管理标准化：Redfish、遥测和固件更新协议.pdf（16页珍藏版）》请在三个皮匠报告上搜索。

1、Choudary Maddukuri(Microsoft)Krishna Sugumaran(AMD)Balaji Vembu(Meta)Qian Wang(NVIDIA)Justin York(Google)GPU Management Interfaces and Firmware UpdateGPU Management Interfaces and Firmware UpdateChoudary Maddukuri(Microsoft)Krishna Sugumaran(AMD)Balaji Vembu(Meta)Qian Wang(NVIDIA)Justin York(Google)

2、HARDWARE MANAGEMENTPanel DiscussionKrishna SugumaranPrincipal Firmware Engineer,MicrosoftChoudary MaddukuriFirmware Architect,AMDBalaji VembuAI Systems Engineer,MetaQian WangSoftware Engineer,NVIDIAJustin YorkSenior Staff Engineer,GoogleOCP UBB Baseline Management RIP v1.0.0*Success:AMD,Google,Micro

3、soft,and NVIDIA use to validate firmware releases for interface compliance OCP GPU Management Interfaces Specification v1.0Success:Location resource used extensively for repair workflowsOCP Firmware Update Requirements for GPUs v1.0Success:Adopted into products for uniform firmware updateDMTF Platfo

4、rm Message Registry v1.3.0(part of Redfish 2025.2)Success:Microsoft using in platform designsGPU Management Interfaces WS Accomplishments*RIP=Redfish Interoperability ProfileAccelerator Universal Base Board(UBB)The GPU management interfaces updates for 2025 focus primarily on efforts to improve mana

5、gement of UBB-form factor accelerator products.These designs have an Accelerator Management Controller performing local management of multiple GPU chips and other base board components.oStandardized GPU Message Registry fields in DMTF Redfish 2025.2 Platform message registry.oFinalizing the GPU Mana

6、gement Interfaces 1.1 publication to address errata and make minor corrections.oWorking with AMD and NVIDIA for supplier agreement on a new Redfish Interoperability Profile,version 1.1,to incorporate additional mandatory properties.oWorking with the DMTF on improving SPDM over Redfish.oTargeting GPU