《HC2022.XeonDx700.PraveenMosur.FINAL.pdf》由会员分享,可在线阅读,更多相关《HC2022.XeonDx700.PraveenMosur.FINAL.pdf(22页珍藏版)》请在三个皮匠报告上搜索。
1、August,2022Built for the Edge:The Next-Generation Intel Xeon D 2700&1700 processors Praveen MosurIntel Fellow Intel ConfidentialDepartment or Event Name2Architected for the Edge Native ApplicationsDesigned for Network,Storage,vRAN,AI and IOT edge workload consolidationIntegrated Accelerators and Eth
2、ernet with Flexible Packet ProcessorOptimized for space&power constrained ruggedized environments Intel Xeon D-2700ProcessorIntel Xeon D-1700ProcessorIntel ConfidentialDepartment or Event Name3Compute OptimizationEdge OptimizationComputeScalar and Data ParallelPCIe Gen 4High Bandwidth IOIntegrated E
3、thernet and AcceleratorsMemoryLow Latency/High BandwidthPerformanceSecurityIntegration and Form FactorScalabilityIntel ConfidentialDepartment or Event NameIntel Xeon D-1700 Processor4=multiple lanesUART,LPC,SPI,eMMC 5.1,USB 2.0Intel Xeon D-2700ProcessorEthernetGen2 QATGen3QAT=multiple lanesDDR452.5m
4、m x 45mm BGAUART,LPC,SPI,eMMC 5.1,USB 2.045mm x 45mm BGA8 Ports Ethernet32 Lanes PCleGen 424 LanesPCIe Gen3,SATA or USB3DDR4DDR4DDR48 Ports Ethernet16 Lanes PCleGen 424 LanesPCIe Gen3,SATA or USB3Gen3 QATEthernet“inline”Intel ConfidentialDepartment or Event Name5Intel XeonD-2100 processor(per core)I
5、ntel Xeon D-2700&D-1700 processors(per core)Out-of-Order Window224352In-Flight Loads+Stores72+56128+72Scheduler Entries97160Register Files Integer+FP180+168280+224Allocation Queue64/thread70/thread;140/1 threadL1D Cache32 KB48 KBL2 Unified TLB(STLB)1.5K2KSTLB-IG Page Support161024(shared w/4K)STLB-I
6、G Page Support161024(shared w/4K)Mid-Level(L2)Cache1 MB1.25 MBImproved Front-End:higher capacity and improved branch predictorWider and deeper machine:wider allocation,larger structures and execution resourcesEnhancements in TLBs,single-thread execution,prefetchingEdge optimized capabilities;larger
7、mid-level cache(L2)+higher vector throughputIntel ConfidentialDepartment or Event Name6Intel Xeon D-2187NTProcessors1Intel Xeon D-2798NXProcessors2Intel Xeon D-2798NXProcessors2(with Vector AES)AES-256 GCM1.01.53x3.4xIntel Xeon D-2187NTProcessors1Intel Xeon D-2798NXProcessors2Intel Xeon D-2798NXProc
8、essors2(with AVX512 IFMA)RSA-2048 Sign1.01.15x4.37xRSA-2048 Verify1.01.14x2.29xECDSA Sign1.01.17x2.13xECDSA Verify1.01.15x1.15xCryptography Vector AES and Vector Carry-Less Multiply Instructions Galois Field New Instructions(GFNI)SHA-NI Big-Number Arithmetic(AVX-512 Integer IFMA)Compression/Decompre
9、ssion and Special SIMD Bit Algebra Vector Bit Manipulation Instructions(VBMI)Cryptography Performance Improvements per core 1 Configuration:1-node,1x Intel Xeon D-2187NT CPU 2.0Ghz on Intel reference platform(Yuba City)with 64 GB(4 slots/16GB/2666)total DDR4 memory,ucode 0 x,HT ON,Turbo OFF,Ubuntu 2
10、0.04 LTS(Focal Fossa),5.4.0-91-generic,OpenSSL 1.1.1,QAT Engine v0.6.10,1x Intel 240G SSD,test by Intel on 2/11/2022.2 Configuration:1-node,1x Intel Xeon D-2798NX CPU 2.1Ghz on Intel reference platform(Moro City)with 64 GB(4 slots/16GB/2933)total DDR4 memory,ucode 0 x1000150,HT ON,Turbo OFF,Ubuntu 2
11、0.04 LTS(Focal Fossa),5.4.0-91-generic,OpenSSL 1.1.1,QAT Engine v0.6.10,1x Intel 240G SSD,test by Intel on 2/11/2022.VIntel ConfidentialDepartment or Event Name7Intel Xeon D-2796NTProcessors1(with AVX512)Intel Xeon D-2796NTProcessors1(with AVX512 VNNI)Resnet-50(images/sec)21.01.8xMobileNet-SSD(image
12、s/sec)31.01.5xVideo Analytics Stream Density41.01.9xEdge AI Growing demand for inferencing and ML at edge Process,manage and analyze data at edge and IoT end-point Reduce latency and bandwidth needIntel Xeon D-2700&D-1700 processors pack AI performance Intel Deep Learning Boost(Intel DL Boost)with V
13、ector Neural Network Instructions(VNNI)Optimized vector multiply-accumulate at 8-bit Efficient inference acceleration1 Refer to Appendix slide for platform configuration and workload details.2 As measured by OpenVINO,ResNet-50,INT8,BS=1,on Intel Xeon D with VNNI enabled vs Intel Xeon D with VNNI dis
14、abled.3 As measured by OpenVINO,SSD-MobileNet,INT8,BS=1,on Intel Xeon D with VNNI enabled vs Intel Xeon D with VNNI disabled.4 As measured by Video Analytics Stream density(1080p30 H.264 video decode,downscaling,color space conversion,object classification w/ResNet50-tf int8on Intel Xeon D with VNNI
15、 enabled vs Intel Xeon D with VNNI disabled.Intel ConfidentialDepartment or Event Name8 Helps provide enhanced security protections for application data independent of operating system or hardware configurationAPPHardwareAPPOperating SystemVirtual Machine ManagerSecure EnclavesIntelSGXData&CodeNo pr
16、oduct or component can be absolutely secure.Minimally-sized Trusted Computing Base helps protect against SW attacks even if OS,drivers,BIOS,or VMM are compromisedHelps increase protections for secrets(data/keys/code IP/etc.)even if attacker has full control of platformHelps prevent attacks such as m
17、emory bus snooping,memory tampering,and“cold boot”attacks against memory contents in RAMIncreases transparency&accountability with option for hardware-based attestation that verifies valid code and data signaturesData&CodeIntel ConfidentialDepartment or Event NameCrypto and compression acceleration
18、functions Bulk Crypto:support for new cryptographic algorithms Chacha20-Poly1305,SM3 and SM4 Public Key Engine Lossless Data Compression:improved compression ratioAdvanced RAS,power management and virtualizationProvide simultaneous acceleration via Inline and Lookaside Host interface for Lookaside p
19、rocessing of requests from host Inline packet interface for IPSec Crypto processing of requests from Ethernet InterfaceHost InterfaceWork Queue ManagerDMAInline Packet InterfaceBulk CryptoPublic Key EngineCompressionHost Interface(Intel QAT Gen 3 available in Intel Xeon D-2700 processors)PacketInter
20、faceConfigurationInterface9Intel ConfidentialDepartment or Event NameCoresIntel QATLook-Aside Modelfor Symmetric&Asymmetric Encryption&Compression/DecompressionIntel Ethernet Controller210CoresIntel QATIntel Ethernet ControllerInline Modelfor IPSec Hardware Acceleration2134512345Intel Con
21、fidentialDepartment or Event NameTransport Layer Security(TLS)Widely used protocol to secure connections between Client&ServersPublic Key Encryption is used to establish symmetric keys between Client&Server Flexible options for accelerating TLS Secure Connection handshakes Software Optimized softwar
22、e using workload accelerator instructions(AVX512 IFMA)Lookaside Acceleration using Gen 3 Intel QuickAssist Technology1.02.54.00.00.51.01.52.02.53.03.54.04.5SoftwareOptimized Software Acceleration withIntel QATRelative Connections Per Second Intel Xeon D-2798NX 4C8TCipher:TLS_AES_128_GCM_SHA256,Curve
23、:X25519,Key:RSA2KHandshakes OnlyNGINX TLS 1.3 Webserver11 Refer to Appendix slide for platform configuration and workload details11Intel ConfidentialDepartment or Event NameIntel Ethernet Controller100 Gbps programmable parsing/classification/modificationRDMA with iWARP and RoCE v2Integrated ACL pro
24、cessingFeature-rich RSS/Flow DirectorAdvanced Scheduling Module:multiple layer hierarchical scheduler with dynamic updates,dual rate shaping,Strict Priority,WFQ or combination schedulingFlexible Packet Processor and SwitchOnly available in Intel Xeon D-2700 processorsSwitch features including port t
25、o port,MAC learning,flexible parsing/classification,policing,ACLs,VM to VMInterfaces with Gen 3 Intel QuickAssist Technology(Intel QAT)to provide Inline IPSec offloadIntel Ethernet ControllerHost Interface8 PortsIntel Xeon D-1700 ProcessorIntel Ethernet ControllerGen 3 Intel QAT(Inline IPSec)Host In
26、terfaceFlexible Packet Processor and Switch8 PortsIntel Xeon D-2700 Processor12Intel ConfidentialDepartment or Event NameCPUCoresLinkEncryption(QAT)PH/MACHot I/FLinkReceive(Ingress)Flow Transmit(Egress)Flow Packet Header&Payload in Clear TextPlaintextEncrypted Packet Header&PayloadIPsecStackParse/Cl
27、assifyWhitelistHit=decrypt or dropParsingClassification in FiltersPacket ModifierHost I/FPHY/MACHost I/FTrafficManagerStateless OffloadsPHY/MACEncryption(QAT)Decryption(QAT)13Intel ConfidentialDepartment or Event NameNetwork security is foundational at the edgeIPSec is used extensively to protect pe
28、er-2-peer network linksFlexible options for IPSec implementationSoftware using Vector AES NILookaside crypto acceleration with IntelQATInline Crypto acceleration with IntelQAT and Integrated Ethernet (Intel Xeon D-2700 processor only)*Configuration:1-node,1x Intel Xeon D-2798NX CPU on Intel referenc
29、e platform(Moro City)with 64 GB(4 slots/16GB/2933)total DDR4 memory,ucode 0 x1000150,HT ON,Turbo OFF,Ubuntu 20.04.3 LTS(Focal Fossa),5.4.0-91-generic,1x Intel 240G SSD,4x25G internal Port,IPSEC DPDK ipsec-secgw application v21.11-force-max-simd-bitwidth=64,Gcc 9.3.0,IpsecMB v1.1,test by Intel on 6/2
30、1/2022.IPSec with Software Encryption with vAES NIIPSec with Lookaside Encryption with Intel QATInline IPSec with Intel QATRelative IPSec Throughput1.5913.7123.5Ethernet Frame Size,Bytes 512B1420BPerformance of DPDK IPSec Security Gateway Intel Single Core Xeon 2798NX ProcessorAES256-GCM Encryption
31、Algorithm14Intel ConfidentialDepartment or Event NameArchitecture choices for the enabling efficient edge computing Optimizing Compute,SoC Fabric Re-targeting Voltage/Frequency operating point based for optimum perf/W based on edge workloadsEssential IO integrated with low power Die-2-Die interfaceO
32、ptimized BGA packages for enabling dense form-factorsExtended temperature support for rugged environments 27012914025Intel Xeon Platinum 8380 ProcessorIntel Xeon Gold 5315Y ProcessorIntel Xeon D-2799ProcessorIntel Xeon D-1702ProcessorIntel Xeon D1700 D&Intel Xeon D27003rdGeneration Intel Xeon Scalab
33、le ProcessorsTDP Scalability of Intel Xeon Architecture 15Intel ConfidentialDepartment or Event NamevRAN:RDU (RU+DU)Radio Unit+Distributed UnitCOM-HPC ModuleFlexible design for Multiple Use Cases 16Intel ConfidentialDepartment or Event NameDesigned from the ground up for the software-defined network
34、 and edgeOptimized for space and power constrained ruggedized environmentsBuilt-in AI and security,integrated crypto acceleration and ethernet with Intel architecture thats known and trusted17Intel Xeon D-2700ProcessorIntel Xeon D-1700ProcessorIntel ConfidentialDepartment or Event NameReady to win?V
35、isit bit.ly/HotWings22 and match Intel speakers to their talks for a chance to win an Intel NUC Mini PC and other prizes.Intel ConfidentialDepartment or Event NameThank you.Intel ConfidentialDepartment or Event NamePerformance varies by use,configuration and other factors.Learn more at www.I results
36、 are based on testing as of dates shown in configurations and may not reflect all publicly available updates.See backup for configuration details.No product or component can be absolutely secure.Your costs and results may vary.Intel technologies may require enabled hardware,software or service activ
37、ation.Intel Corporation.Intel,the Intel logo,and other Intel marks are trademarks of Intel Corporation or its subsidiaries.Other names and brands may be claimed as the property of others.20Intel ConfidentialDepartment or Event NamePerformance varies by use,configuration and other factors.Learn more
38、at www.I Xeon D-2798NX:1-node,1x Intel Xeon D-2798NX CPU 2.1Ghz on Intel reference platform(Moro City)with 64 GB(4 slots/16GB/2933)total DDR4 memory,ucode 0 x1000150,HT ON,Turbo OFF,Ubuntu 20.04 LTS(Focal Fossa),5.4.0-67-generic,NGINX v0.4.7,OpenSSL 1.1.1l,1x Intel 240G SSD,test by Intel on 2/28/202
39、2.Intel Xeon D-2798NX Optimized SW:1-node,1x Intel Xeon D-2798NX CPU 2.1Ghz on Intel reference platform(Moro City)with 64 GB(4 slots/16GB/2933)total DDR4 memory,ucode 0 x1000150,HT ON,Turbo OFF,Ubuntu 20.04 LTS(Focal Fossa),5.4.0-67-generic,NGINX v0.4.7,OpenSSL 1.1.1l,QAT Engine v0.6.10,Intel IPsec
40、MB v1.1,IPP-Crypto ippcp_2021.4,1x Intel 240G SSD,test by Intel on 2/28/2022.Intel Xeon D-2798NX QAT Accelerated:1-node,1x Intel Xeon D-2798NX CPU 2.1Ghz on Intel reference platform(Moro City)with 64 GB(4 slots/16GB/2933)total DDR4 memory,ucode 0 x1000150,HT ON,Turbo OFF,Ubuntu 20.04 LTS(Focal Fossa
41、),5.4.0-67-generic,NGINX v0.4.7,OpenSSL 1.1.1l,QAT18.L.1.4.0-00008,1x Intel 240G SSD,test by Intel on 2/28/2022.*Other names and brands may be claimed as the property of others21Intel ConfidentialDepartment or Event NameHostNamemorocity8-24-3-3TimeThu Jul 21 06:07:02 PM UTC 2022SystemManufacturerACC
42、TONProduct NameMOROCITYVersion1.2.2.1Serial#MR400411-1123-785-980-11042UUIDa5a5a5a5-a5a5-0862-0112-150fa5a5a5a5BaseboardManufacturerACCTONProduct NameMOROCITYVersionK88661-101Serial#MR400411-1123-785-980-42ChassisManufacturerIntel CorporationTypeRack Mount ChassisVersion8675-309-401-412Serial#SN0009
43、UDBIOSVendorIntel CorporationVersionIDVLCRB1.86B.0021.D41.2112031014Release Date12/03/2021Operating SystemOSUbuntu 22.04 LTSKernel5.15.0-27-genericMicrocode0 x1000150Software GCCgcc(Ubuntu 11.2.0-19ubuntu1)11.2.0GLIBCldd(Ubuntu GLIBC 2.35-0ubuntu3)2.35BinutilsGNU ld(GNU Binutils for Ubuntu)2.38Pytho
44、nPython 3.9.12Python3Python 3.9.12JavaOpenSSLOpenSSL 1.1.1n 15 Mar 2022CPUCPU ModelIntel(R)Xeon(R)D-2796NT CPU 2.00GHzArchitecturex86_64MicroarchitectureICXFamily6Model108Stepping1Base Frequency 2.0GHzMaximum Frequency2.0GHzAll-core Maximum Frequency2.5GHzCPUs40On-line CPU List 0-39Hyperthreading EnabledCores per Socket20Sockets1NUMA Nodes1NUMA CPU List 0-39CHA Count12L1d Cache960 KiB(20 instances)L1i Cache640 KiB(20 instances)L2 Cache25 MiB(20 instances)L3 Cache30 MiB(1 instance)Memory Channels8PrefetchersL2 HW,L2 Adj.,DCU HW,DCU IPIntel Turbo BoostEnabledPPINs207a4f4d6aad8f5922