当前位置:首页 > 报告详情

记忆助推建筑业巨大变革.pdf

上传人: a****d 编号:185007 2024-10-07 37页 1.69MB

1、1|2024 SNIA.All Rights Reserved.Memories are Driving Big Architectural Changes.Hold Onto Your Hats!Presented byJim Handy,Objective AnalysisTom Coughlin,Coughlin Associates,President,IEEE2|2024 SNIA.All Rights Reserved.OutlineChange 1:Emerging Memories ArriveChange 2:Memory DisaggregationChange 3:Pro

2、cessing in MemoryChange 4:ChipletsChange 5:AI Everywhere3|2024 SNIA.All Rights Reserved.OutlineChange 1:Emerging Memories ArriveChange 2:Memory DisaggregationChange 3:Processing in MemoryChange 4:ChipletsChange 5:AI Everywhere4|2024 SNIA.All Rights Reserved.Emerging Memories Bear Strong Similarities

3、Graphic courtesy of Dr.Johannes Mller5|2024 SNIA.All Rights Reserved.What is an“Emerging Memory?”Established Memory Large market:$1B+Worrisome scaling limit Well-understood process Based on silicon alone You know these names:DRAM NAND flash NOR flash SRAM Specialty(FIFO,CAM)Emerging Memory Small Mar

4、ket:0.9910 X fasterIntel Xeon Gold GSI Gem1050100150200250300350SERVER POWER IN WATTS2 X Lower system power Idle APUCPU25|2024 SNIA.All Rights Reserved.Analog Neural NetsOld idea seeing renewed interestInstant Matrix Algebra Single cycle math for entire matrix multiplySimple operation Difficult to s

5、et upA good accelerator to a standard CPUFits emerging memories wellLots of research,but few products26|2024 SNIA.All Rights Reserved.One or More of these May SucceedProspective co-processor to offload tasks from a standard CPUPossibly the main processor in smaller edge applications Low-end CPU or M

6、PU will perform housekeepingCould slow CPU performance increases CPU only needed for less significant&slower tasks More complicated tasks offloaded to the PIM chip Overall cost savings resultPerformance scales with the addition of PIM chips One CPU per 10/100/1,000 PIMs27|2024 SNIA.All Rights Reserv

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。
标记中的内容概述了新兴存储技术对计算架构的重大影响。文章指出,新兴存储技术如MRAM、ReRAM和PCM等,具有非易失性,并且有望在规模和性能上实现显著提升。预计到2033年,这些技术的收入将大幅增长,远超现有的DRAM和NAND闪存。新兴存储技术对计算的影响包括将持久性更接近处理器,减少数据移动,以及降低数据中心能耗。文章还讨论了内存分解、内存中的处理、芯片小片化和人工智能的普及等趋势。内存分解将消除“孤立内存”,而芯片小片化则通过标准化接口,降低了成本和提高了性能。最后,文章强调了人工智能在提高响应速度、减少带宽需求、提升数据完整性、增强安全性和改善用户体验方面的潜力。总体而言,这些技术的发展将为计算领域带来深刻变革,需要我们“握紧帽子”。
"新兴记忆技术将如何改变计算架构?" "内存分解如何提升数据中心的效率?" "在内存中处理数据有哪些优势和挑战?"
客服
商务合作
小程序
服务号
折叠