当前位置:首页 > 报告详情

扩展开放计算:RISC-V、Chiplets 以及人工智能和机器人技术的未来.pdf

上传人: c** 编号:955305 2025-10-27 20页 9.82MB

1、Scaling Open Compute:RISC-V,Chiplets,and the Future of AI and RoboticsWei-Han Lien Chief Architect 4Golden Age of Silicon DiversitySiliconSilicon HeterogeneityHeterogeneityArchitecture Architecture V Variety ariety MultiMulti-vendor vendor Silicon Silicon L Landscape andscape Design Design Compossib

2、ility Compossibility AI PersonalizationUbiquitous AI Compute Open Source Hardware 5ExtensibleScalableEfficientStable6L2TLBIntPRFreelistTWFill Data 64BTW AddressMMUHistory File VF PRFreelistFP PRFreelistCheck-PointsMapperRetireQueueDB0ALUMULFPALUDIVFPALUALURS0RS1RS2 RS3INT/FP RFDB1LD/ST2LD2RS6RS7VFP

3、RFRS4RS5DB3DB4Fill Data 64B Miss/Write AddressWrite64B DataInteger Cluster Unit 0DCTAGDTLBDCACHE256KBx8WDC Way-predictorLDQSTQPattern/stride PrefetcherAddressQueue Write DataQueue FillBufferLoad/Store UnitINT PRFreelistICTAGITLBICACHEIC Way-predictorInstruction Cache Complex Instruction Address Plan

4、eInstruction Data Plane32B FetchI-Miss AddressFill Data 64BPrefetchTAGEBTPTAGEBDPRASNFPBranch Prediction ComplexBranch State TableAlignerFetch BufferDEC13I1I0DEC2DEC1DEC0I2I3I4I5DEC14DEC15I6I7Uop3Uop2Uop1Uop0Uop13Uop14Uop15RS2 RS3DB2BRBRINT RFRS4BTBUOPCacheuCache Uop MaxVFP00-3Integer Cluster Unit 1

5、ST2Foundational Compute IPsSOC TechnologyChipletTechnologyCPUPCIeNetworkingMemoryFabricPower/Thermal ManagementRASSecurityDebugCPU,AI,IO,Memory chipletsApplicationsData CenterTenstorrentTensixNeo CoreTensixNeo CoreTensixNeo CoreTensixNeo CoreL1OverlayNoCTensix NEO YayuiAIDCRISC-V CPU Diversity9RISC-

6、V in the Cross-road RISC-Vs 15-year Journey An academic project in UC Berkeley Mainstream adoption in embedded and industrial applications Sustain momentum High-performance computing implementation Software development Key Requirements Advanced O-o-O CPU design with high-bandwidth memory Optimized c

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《Scaling Open Compute: RISC-V, Chiplets, and the Future of AI and Robotics》的内容,以下是全文关键点的概括: 1. **硅多样性时代**:进入硅异构、架构多样、多厂商的硅景观时代。 2. **开放硬件**:强调可扩展、可伸缩、高效、稳定的开源硬件。 3. **RISC-V CPU**:RISC-V处理器在学术和工业应用中逐渐被主流接受,具有高性能计算实现和软件开发优势。 4. **高性能处理器**:如Ascalon CPU集群,具有18 SPECINT/GHz性能,支持PCIe Gen6x16和LPDDR5X。 5. **芯片片技术**:采用芯片片技术,如Ascalon-Auto IP,实现ISO26262功能安全特性。 6. **AI和机器人**:结合神经符号AI芯片,优化CPU/GPU数据传输延迟,推动AI个性化与普及。 7. **开放芯片架构(OCA)**:通过OCA实现芯片片互操作性,促进不同组织实体间的芯片片多样性。
未来AI计算的关键?" OCA如何定义未来?" AI芯片如何引领变革?"
客服
商务合作
小程序
服务号
折叠