Intel Launches 4th Gen Xeon Scalable Processors, Max Series CPUs

Intel high짯lights broad indus짯try adop짯ti짯on across all major CSPs, OEMs, ODMs and ISVs, and show짯ca짯ses increased per짯for짯mance in AI, net짯wor짯king and high per짯for짯mance computing.

NEWS HIGHLIGHTS
 

  • Expan짯si짯ve cus짯to짯mer and part짯ner adop짯ti짯on from AWS, Cis짯co, Clou짯de짯ra, Core짯Wea짯ve, Dell Tech짯no짯lo짯gies, Drop짯box, Erics짯son, Fuji짯tsu, Goog짯le Cloud, Hew짯lett Packard Enter짯pri짯se, IBM Cloud, Inspur Infor짯ma짯ti짯on, IONOS, Leno짯vo, Los Ala짯mos Natio짯nal Labo짯ra짯to짯ry, Micro짯soft Azu짯re, NVIDIA, Ora짯cle Cloud, OVH짯cloud, phoe짯nix짯NAP, Red짯Hat, SAP, Super짯Mi짯cro, Tele짯fo짯ni짯ca and VMware, among others.
  • With the most built-in acce짯le짯ra짯tors of any CPU in the world for key workloads such as AI, ana짯ly짯tics, net짯wor짯king, secu짯ri짯ty, sto짯rage and high per짯for짯mance com짯pu짯ting (HPC), 4th Gen Intel Xeon Sca짯lable and Intel Max Series fami짯lies deli짯ver lea짯der짯ship per짯for짯mance in a pur짯po짯se-built workload-first approach.
  • 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors are Intel셲 most sus짯tainable data cen짯ter pro짯ces짯sors, deli짯ve짯ring a ran짯ge of fea짯tures for opti짯mi짯zing power and per짯for짯mance, making opti짯mal use of CPU resour짯ces to help achie짯ve cus짯to짯mers sus짯taina짯bi짯li짯ty goals.
  • When com짯pared with pri짯or gene짯ra짯ti짯ons, 4th Gen Xeon cus짯to짯mers can expect a 2.9x1 avera짯ge per짯for짯mance per watt effi짯ci짯en짯cy impro짯ve짯ment for tar짯ge짯ted workloads when uti짯li짯zing built-in acce짯le짯ra짯tors, up to 70-watt2 power savings per CPU in opti짯mi짯zed power mode with mini짯mal per짯for짯mance loss for sel짯ect workloads and a 52% to 66% lower total cost of owner짯ship (TCO)3.

SANTA CLARA, Calif., Jan. 10, 2023 Intel today mark짯ed one of the most important pro짯duct laun짯ches in com짯pa짯ny histo짯ry with the unvei짯ling of 4th Gen Intel짰 Xeon짰 Sca짯lable pro짯ces짯sors (code-named Sap짯phi짯re Rapids), the Intel짰 Xeon짰 CPU Max Series (code-named Sap짯phi짯re Rapids HBM) and the Intel짰 Data Cen짯ter GPU Max Series (code-named Pon짯te Vec짯chio), deli짯ve짯ring for its cus짯to짯mers a leap in data cen짯ter per짯for짯mance, effi짯ci짯en짯cy, secu짯ri짯ty and new capa짯bi짯li짯ties for AI, the cloud, the net짯work and edge, and the world셲 most powerful supercomputers.

Working along짯side its cus짯to짯mers and part짯ners with 4th Gen Xeon, Intel is deli짯ve짯ring dif짯fe짯ren짯tia짯ted solu짯ti짯ons and sys짯tems at sca짯le to tack짯le their big짯gest com짯pu짯ting chal짯lenges. Intel셲 uni짯que approach to pro짯vi짯ding pur짯po짯se-built, workload-first acce짯le짯ra짯ti짯on and high짯ly opti짯mi짯zed soft짯ware tun짯ed for spe짯ci짯fic workloads enables the com짯pa짯ny to deli짯ver the right per짯for짯mance at the right power for opti짯mal over짯all total cost of ownership.

Press Kit: 4th Gen Xeon Sca짯lable Processors

Addi짯tio짯nal짯ly, as Intel셲 most sus짯tainable data cen짯ter pro짯ces짯sors, 4th Gen Xeon pro짯ces짯sors deli짯ver cus짯to짯mers a ran짯ge of fea짯tures for mana짯ging power and per짯for짯mance, making the opti짯mal use of CPU resour짯ces to help achie짯ve their sus짯taina짯bi짯li짯ty goals.

The launch of 4th Gen Xeon Sca짯lable pro짯ces짯sors and the Max Series pro짯duct fami짯ly is a pivo짯tal moment in fue짯ling Intel셲 tur짯n짯around, reig짯ni짯ting our path to lea짯der짯ship in the data cen짯ter and gro짯wing our foot짯print in new are짯nas, said San짯dra Rive짯ra, Intel exe짯cu짯ti짯ve vice pre짯si짯dent and gene짯ral mana짯ger of the Data Cen짯ter and AI Group. 쏧ntel셲 4th Gen Xeon and the Max Series pro짯duct fami짯ly deli짯ver what cus짯to짯mers tru짯ly want lea짯der짯ship per짯for짯mance and relia짯bi짯li짯ty within a secu짯re envi짯ron짯ment for their real-world requi짯re짯ments dri짯ving fas짯ter time to value and powe짯ring their pace of innovation.

Unli짯ke any other data cen짯ter pro짯ces짯sor on the mar짯ket and alre짯a짯dy in the hands of cus짯to짯mers today, the 4th Gen Xeon fami짯ly great짯ly expands on Intel셲 pur짯po짯se-built, workload-first stra짯tegy and approach. 

Lea짯ding Per짯for짯mance and Sus짯taina짯bi짯li짯ty Bene짯fits with the Most Built-In Acceleration

Today, the짯re are over 100 mil짯li짯on Xeons instal짯led in the mar짯ket from on-prem ser짯vers run짯ning IT ser짯vices, inclu짯ding new as-a-ser짯vice busi짯ness models, to net짯wor짯king equip짯ment mana짯ging Inter짯net traf짯fic, to wire짯less base sta짯ti짯on com짯pu짯ting at the edge, to cloud services.

Buil짯ding on deca짯des of data cen짯ter, net짯work and intel짯li짯gent edge inno짯va짯ti짯on and lea짯der짯ship, new 4th Gen Xeon pro짯ces짯sors deli짯ver lea짯ding per짯for짯mance with the most built-in acce짯le짯ra짯tors of any CPU in the world to tack짯le cus짯to짯mers most important com짯pu짯ting chal짯lenges across AI, ana짯ly짯tics, net짯wor짯king, secu짯ri짯ty, sto짯rage and HPC.

 

When com짯pa짯ring with pri짯or gene짯ra짯ti짯ons, 4th Gen Intel Xeon cus짯to짯mers can expect a 2.9x1 avera짯ge per짯for짯mance per watt effi짯ci짯en짯cy impro짯ve짯ment for tar짯ge짯ted workloads when uti짯li짯zing built-in acce짯le짯ra짯tors, up to 70-watt2 power savings per CPU in opti짯mi짯zed power mode with mini짯mal per짯for짯mance loss, and a 52% to 66% lower TCO3.

Sus짯taina짯bi짯li짯ty

The expan짯si짯ve짯ness of built-in acce짯le짯ra짯tors included in 4th Gen Xeon means Intel deli짯vers plat짯form-level power savings, les짯sening the need for addi짯tio짯nal dis짯crete acce짯le짯ra짯ti짯on and hel짯ping our cus짯to짯mers achie짯ve their sus짯taina짯bi짯li짯ty goals. Addi짯tio짯nal짯ly, the new Opti짯mi짯zed Power Mode can deli짯ver up to 20% socket power savings with a less than 5% per짯for짯mance impact for sel짯ec짯ted workloads11. New inno짯va짯tions in air and liquid coo짯ling redu짯ce total data cen짯ter ener짯gy con짯sump짯ti짯on fur짯ther; and for the manu짯fac짯tu짯ring of 4th Gen Xeon, it셲 been built with  90% or more rene짯wa짯ble elec짯tri짯ci짯ty at Intel sites with sta짯te-of-the-art water recla짯ma짯ti짯on facilities.

Arti짯fi짯ci짯al Intelligence

In AI, and com짯pared to pre짯vious gene짯ra짯ti짯on, 4th Gen Xeon pro짯ces짯sors achie짯ve up to 10x5,6 hig짯her PyTorch real-time infe짯rence and trai짯ning per짯for짯mance with built-in Intel짰 Advan짯ced Matrix Exten짯si짯on (Intel짰 AMX) acce짯le짯ra짯tors. Intel셲 4th Gen Xeon unlocks new levels of per짯for짯mance for infe짯rence and trai짯ning across a wide breadth of AI workloads. The Xeon CPU Max Series expands on the짯se capa짯bi짯li짯ties for natu짯ral lan짯guage pro짯ces짯sing, with cus짯to짯mers see짯ing up to a 20x12 speed-up on lar짯ge lan짯guage models. With the deli짯very of Intel셲 AI soft짯ware suite, deve짯lo짯pers can use their AI tool of choice, while incre짯asing pro짯duc짯ti짯vi짯ty and spee짯ding time to AI deve짯lo짯p짯ment. The suite is por짯ta짯ble from the work짯sta짯tion, enab짯ling it to sca짯le out in the cloud and all the way out to the edge. And it has been vali짯da짯ted with over 400 machi짯ne lear짯ning and deep lear짯ning AI models across the most com짯mon AI uses cases in every busi짯ness segment. 

Net짯wor짯king

4th Gen Xeon offers a fami짯ly of pro짯ces짯sors spe짯ci짯fi짯cal짯ly opti짯mi짯zed for high-per짯for짯mance, low-laten짯cy net짯work and edge workloads. The짯se pro짯ces짯sors are a cri짯ti짯cal part of the foun짯da짯ti짯on dri짯ving a more soft짯ware-defi짯ned future for indus짯tries ran짯ging from tele짯com짯mu짯ni짯ca짯ti짯ons and retail to manu짯fac짯tu짯ring and smart cities. For 5G core workloads, built-in acce짯le짯ra짯tors help increase through짯put and decrease laten짯cy, while advan짯ces in power manage짯ment enhan짯ce both the respon짯si짯ve짯ness and the effi짯ci짯en짯cy of the plat짯form. And, when com짯pared to pre짯vious gene짯ra짯ti짯ons, 4th Gen Xeon deli짯vers up to twice the vir짯tua짯li짯zed radio access net짯work (vRAN) capa짯ci짯ty wit짯hout incre짯asing power con짯sump짯ti짯on. This enables com짯mu짯ni짯ca짯ti짯ons ser짯vice pro짯vi짯ders to dou짯ble the per짯for짯mance-per-watt to meet their cri짯ti짯cal per짯for짯mance, sca짯ling and ener짯gy effi짯ci짯en짯cy needs.

High Per짯for짯mance Computing

4th Gen Xeon and the Intel Max Series pro짯duct fami짯ly bring a sca짯lable, balan짯ced archi짯tec짯tu짯re that inte짯gra짯tes CPU and GPU with oneAPI셲 open soft짯ware eco짯sys짯tem for deman짯ding com짯pu짯ting workloads in HPC and AI, sol짯ving the world셲 most chal짯len짯ging problems.

The Xeon CPU Max Series is the first and only x86-based pro짯ces짯sor with high band짯width memo짯ry, acce짯le짯ra짯ting many HPC workloads wit짯hout the need for code chan짯ges. The Intel Data Cen짯ter GPU Max Series is Intel셲 hig짯hest-den짯si짯ty pro짯ces짯sor and will be available in seve짯ral form fac짯tors that address dif짯fe짯rent cus짯to짯mer needs.

The Xeon CPU Max Series offers 64 giga짯bytes of high band짯width memo짯ry (HBM2e) on the packa짯ge, signi짯fi짯cant짯ly incre짯asing data through짯put for HPC and AI workloads. Com짯pared with top-end 3rd Gen Intel짰 Xeon짰 Sca짯lable pro짯ces짯sors, the Xeon CPU Max Series pro짯vi짯des up to 3.7 times10 more per짯for짯mance on a ran짯ge of real-world appli짯ca짯ti짯ons like ener짯gy and earth sys짯tems modeling.

Fur짯ther, the Data Cen짯ter GPU Max Series packs over 100 bil짯li짯on tran짯sis짯tors into a 47-tile packa짯ge, brin짯ging new levels of through짯put to chal짯len짯ging workloads like phy짯sics, finan짯cial ser짯vices and life sci짯en짯ces. When pai짯red with the Xeon CPU Max Series, the com짯bi짯ned plat짯form achie짯ves up to 12.8 times13 grea짯ter per짯for짯mance than the pri짯or gene짯ra짯ti짯on when run짯ning the LAMMPS mole짯cu짯lar dyna짯mics simulator.

Most Fea짯ture-Rich and Secu짯re Xeon Plat짯form Yet

Signi짯fy짯ing the big짯gest plat짯form trans짯for짯ma짯ti짯on Intel has deli짯ver짯ed, not only is 4th Gen Xeon a mar짯vel of acce짯le짯ra짯ti짯on, but it is also an achie짯ve짯ment in manu짯fac짯tu짯ring, com짯bi짯ning up to four Intel 7멳uilt tiles on a sin짯gle packa짯ge, con짯nec짯ted using Intel EMIB (embedded mul짯ti-die inter짯con짯nect bridge) pack짯a짯ging tech짯no짯lo짯gy and deli짯ve짯ring new fea짯tures inclu짯ding increased memo짯ry band짯width with DDR5, increased I/O band짯width with PCIe5.0 and Com짯pu짯te Express Link (CXL) 1.1 interconnect. 

At the foun짯da짯ti짯on of it all is secu짯ri짯ty. With 4th Gen Xeon, Intel is deli짯ve짯ring the most com짯pre짯hen짯si짯ve con짯fi짯den짯ti짯al com짯pu짯ting port짯fo짯lio of any data cen짯ter sili짯con pro짯vi짯der in the indus짯try, enhan짯cing data secu짯ri짯ty, regu짯la짯to짯ry com짯pli짯ance and data sove짯reig짯n짯ty. Intel remains the only sili짯con pro짯vi짯der to offer appli짯ca짯ti짯on iso짯la짯ti짯on for data cen짯ter com짯pu짯ting with Intel짰 Soft짯ware Guard Exten짯si짯ons (Intel짰 SGX), which pro짯vi짯des today셲 smal짯lest attack sur짯face for con짯fi짯den짯ti짯al com짯pu짯ting in pri짯va짯te, public and cloud-to-edge envi짯ron짯ments. Addi짯tio짯nal짯ly, Intel셲 new vir짯tu짯al-machi짯ne (VM) iso짯la짯ti짯on tech짯no짯lo짯gy, Intel짰 Trust Domain Exten짯si짯ons (Intel짰 TDX), is ide짯al for port짯ing exis짯ting appli짯ca짯ti짯ons into a con짯fi짯den짯ti짯al envi짯ron짯ment and will debut with Micro짯soft Azu짯re, Ali짯baba Cloud, Goog짯le Cloud and IBM Cloud.

Final짯ly, the modu짯lar archi짯tec짯tu짯re of 4th Gen Xeon allows Intel to offer a wide ran짯ge of pro짯ces짯sors across near짯ly 50 tar짯ge짯ted SKUs for cus짯to짯mer use cases or appli짯ca짯ti짯ons, from main짯stream gene짯ral-pur짯po짯se SKUs to pur짯po짯se-built SKUs for cloud, data짯ba짯se and ana짯ly짯tics, net짯wor짯king, sto짯rage, and sin짯gle-socket edge use cases. The 4th Gen Xeon pro짯ces짯sor fami짯ly is On Demand-capa짯ble and varies in core count, fre짯quen짯cy, mix of acce짯le짯ra짯tors, power enve짯lo짯pe and memo짯ry through짯put as is appro짯pria짯te for tar짯get use cases and form fac짯tors addres짯sing cus짯to짯mers real-world requirements.

SKU TABLE: SKUs for 4th Gen Xeon and Intel Xeon CPU Max Series

 

쨔 Geo짯me짯an of fol짯lo짯wing workloads:  Rock짯sDB (IAA vs ZTD), Click짯House (IAA vs ZTD), SPDK lar짯ge media and data짯ba짯se request pro짯xies (DSA vs out of box), Image Clas짯si짯fi짯ca짯ti짯on Res짯Net-50 (AMX vs VNNI), Object Detec짯tion SSD-Res짯Net-34 (AMX vs VNNI), QAT짯zip (QAT vs zlib)

짼 1몁ode, Intel Refe짯rence Vali짯da짯ti짯on Plat짯form, 2x Intel짰 Xeon 8480+ (56C, 2GHz, 350W TDP), HT On,  Tur짯bo ON,  Total Memo짯ry: 1 TB (16 slots/ 64GB/ 4800 MHz), 1x P4510 3.84TB NVMe PCIe Gen4 dri짯ve, BIOS: 0091.D05, (ucode:0x2b0000c0),  Cent짯OS Stream 8, 5.15.0몊pr.bkc.pc.10.4.11.x86_64, Java Perf/Watt w/ openjdk-11+28_linux-x64_bin, 112 ins짯tances, 1550MB Initial/Max heap size, Tes짯ted by Intel as of Oct 2022.

쨀 ResNet50 Image Classification

New Con짯fi짯gu짯ra짯ti짯on: 1몁ode, 2x pre-pro짯duc짯tion 4th Gen Intel짰 Xeon짰 Sca짯lable 8490H pro짯ces짯sor (60 core) with Intel짰 Advan짯ced Matrix Exten짯si짯ons (Intel AMX),  on pre-pro짯duc짯tion Super짯Mi짯cro SYS-221H-TNR with 1024GB DDR5 memo짯ry (16x64 GB), micro짯code 0x2b0000c0, HT On, Tur짯bo On, SNC Off, Cent짯OS Stream 8, 5.19.16301.fc37.x86_64, 1x3.84TB P5510 NVMe, 10GbE x540-AT2, Intel TF 2.10, AI Model=Resnet 50 v1_5, best scores achie짯ved: BS1 AMX 1 core/instance (max. 15ms SLA),  using phy짯si짯cal cores, tes짯ted by Intel Novem짯ber  2022. Base짯line: 1몁ode, 2x pro짯duc짯tion 3rd Gen Intel Xeon Sca짯lable 8380 Pro짯ces짯sor ( 40 cores) on Super짯Mi짯cro SYS-220U-TNR , DDR4 memo짯ry total 1024GB (16x64 GB), micro짯code 0xd000375, HT On, Tur짯bo On, SNC Off, Cent짯OS Stream 8, 5.19.16301.fc37.x86_64, 1x3.84TB P5510 NVMe, 10GbE x540-AT2, Intel TF 2.10, AI Model=Resnet 50 v1_5, best scores achie짯ved: BS1 INT8 2 cores/instance (max. 15ms SLA), using phy짯si짯cal cores, tes짯ted by Intel Novem짯ber 2022.

For a 50 ser짯ver fleet of 3rd Gen Xeon 8380 (RN50 w/DLBoost), esti짯ma짯ted as of Novem짯ber 2022: 

CapEx cos짯ts: $1.64M

OpEx cos짯ts (4 year, includes power and coo짯ling uti짯li짯ty cos짯ts, infra짯struc짯tu짯re and hard짯ware main짯ten짯an짯ce cos짯ts): $739.9K

Ener짯gy use in kWh (4 year, per ser짯ver): 44627, PUE 1.6

Other assump짯ti짯ons:  uti짯li짯ty cost $0.1/kWh,  kWh to kg CO2 fac짯tor 0.42394 

For a 17 ser짯ver fleet of 4th Gen Xeon 8490H (RN50 w/AMX), esti짯ma짯ted as of Novem짯ber 2022: 

CapEx cos짯ts: $799.4K

OpEx cos짯ts (4 year, includes power and coo짯ling uti짯li짯ty cos짯ts, infra짯struc짯tu짯re and hard짯ware main짯ten짯an짯ce cos짯ts): $275.3K

Ener짯gy use in kWh (4 year, per ser짯ver): 58581, PUE 1.6

AI 55% lower TCO by deploy짯ing fewer 4th Gen Intel짰 Xeon짰 pro짯ces짯sor-based ser짯vers to meet the same per짯for짯mance requi짯re짯ment. See [E7] at intel.com/processorclaims: 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors. Results may vary.

Data짯ba짯se 52% lower TCO by deploy짯ing fewer 4th Gen Intel짰 Xeon짰 pro짯ces짯sor-based ser짯vers to meet the same per짯for짯mance requi짯re짯ment. See [E8] at intel.com/processorclaims: 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors. Results may vary.

HPC 66% lower TCO by deploy짯ing fewer Intel짰 Xeon짰 CPU Max pro짯ces짯sor-based ser짯vers to meet the same per짯for짯mance requi짯re짯ment. See [E9] at intel.com/processorclaims: 4th Gen Intel Xeon Sca짯lable pro짯ces짯sors. Results may vary.

4 Geo짯me짯an of HP Lin짯pack, Stream Tri짯ad, SPECrate2017_fp_base est, SPECrate2017_int_base est. See [G2, G4, G6] at intel.com/processorclaims: 4th Gen Intel Xeon Scalable.

5 Up to 10x hig짯her PyTorch real-time infe짯rence per짯for짯mance with built-in Intel짰 Advan짯ced Matrix Exten짯si짯ons (Intel짰 AMX) (BF16) vs. the pri짯or gene짯ra짯ti짯on (FP32)

PyTorch geo짯me짯an of ResNet50, Bert-Lar짯ge, Mas짯kRCNN, SSD-Res짯Ne짯t34, RNN멦, Resnext101.

6 Up to 10x hig짯her PyTorch trai짯ning per짯for짯mance with built-in Intel짰 Advan짯ced Matrix Exten짯si짯ons (Intel짰 AMX) (BF16) vs. the pri짯or gene짯ra짯ti짯on (FP32)

PyTorch geo짯me짯an of ResNet50, Bert-Lar짯ge, DLRM, Mas짯kRCNN, SSD-Res짯Ne짯t34, RNN멦.

7 Esti짯ma짯ted as of 8/30/2022 based on 4th gene짯ra짯ti짯on Intel짰 Xeon짰 Sca짯lable pro짯ces짯sor archi짯tec짯tu짯re impro짯ve짯ments vs 3rd gene짯ra짯ti짯on Intel짰 Xeon짰 Sca짯lable pro짯ces짯sor at simi짯lar core count, socket power and fre짯quen짯cy on a test sce짯na짯rio using Flex짯RAN꽓 soft짯ware. Results may vary.

8 Up to 95% fewer cores and 2x hig짯her level 1 com짯pres짯si짯on through짯put with 4th Gen Intel Xeon Pla짯ti짯num 8490H using inte짯gra짯ted Intel QAT vs. pri짯or generation.

8490H: 1몁ode, pre-pro짯duc짯tion plat짯form with 2x 4th  Gen Intel짰 Xeon Sca짯lable Pro짯ces짯sor (60 core) with inte짯gra짯ted Intel Quick짯As짯sist Acce짯le짯ra짯tor (Intel QAT), QAT device utilized=8(2 sockets acti짯ve), with Total 1024GB (16x64 GB) DDR5 memo짯ry, micro짯code 0xf000380, HT On, Tur짯bo Off, SNC Off, Ubun짯tu 22.04.1 LTS, 5.15.047-generic, 1x 1.92TB Intel짰 SSDSC2KG01, QAT v20.l.0.9.1 , QAT짯zip v1.0.9 , ISA멛 v2.3.0, tes짯ted by Intel Sep짯tem짯ber 2022. 

8380: 1몁ode, 2x 3rd Gen Intel Xeon Sca짯lable Pro짯ces짯sors( 40 cores) on Coyo짯te Pass plat짯form, DDR4 memo짯ry total 1024GB (16x64 GB), micro짯code 0xd000375, HT On, Tur짯bo Off, SNC Off, Ubun짯tu 22.04.1 LTS, 5.15.047-generic, 1x 1.92TB Intel SSDSC2KG01,QAT v1.7.l.4.16,  QAT짯zip v1.0.9 , ISA멛 v2.3.0, tes짯ted by Intel Octo짯ber 2022.

9 Up to 3x hig짯her Rock짯sDB per짯for짯mance with 4th Gen Intel Xeon Pla짯ti짯num 8490H using inte짯gra짯ted Intel IAA vs. pri짯or generation.

8490H: 1몁ode, pre-pro짯duc짯tion Intel plat짯form with 2x 4th Gen Intel Xeon Sca짯lable Pro짯ces짯sor (60 cores) with inte짯gra짯ted Intel In-Memo짯ry Ana짯ly짯tics Acce짯le짯ra짯tor (Intel IAA), HT On, Tur짯bo On, Total Memo짯ry 1024GB (16x64GB DDR5 4800),  micro짯code 0xf000380,  1x 1.92TB INTEL SSDSC2KG01, Ubun짯tu 22.04.1 LTS, 5.18.12051812-generic, QPL v0.1.21,accel-config-v3.4.6.4, ZSTD v1.5.2, Rock짯sDB v6.4.6 (db_bench), tes짯ted by Intel  Sep짯tem짯ber 2022.

8380: 1몁ode, 2x  3rd Gen Intel Xeon Sca짯lable Pro짯ces짯sors( 40 cores) on Coyo짯te Pass plat짯form, HT On, Tur짯bo On, SNC Off, Total Memo짯ry 1024GB (16x64GB DDR4 3200), micro짯code 0xd000375, 1x 1.92TB INTEL SSDSC2KG01, Ubun짯tu 22.04.1 LTS, 5.18.12051812-generic,  ZSTD v1.5.2, Rock짯sDB v6.4.6 (db_bench), tes짯ted by Intel  Octo짯ber 2022.

10 Intel짰 Xeon짰 8380: Test by Intel as of 10/7/2022. 1몁ode, 2x Intel짰 Xeon짰 8380 CPU, HT On, Tur짯bo On, Total Memo짯ry 256 GB (16x16GB 3200MT/s DDR4), BIOS Ver짯si짯on SE5C620.86B.01.01.0006.2207150335, ucode revision=0xd000375, Rocky Linux 8.6, Linux ver짯si짯on 4.18.0372.26.1.el8_6.crt1.x86_64, YASK v3.05.07

Intel짰 Xeon짰 CPU Max Series: Test by Intel as of ww3622. 1몁ode, 2x Intel짰 Xeon짰 CPU Max SeriesHT On, Tur짯bo On, SNC4, Total Memo짯ry 128 GB (8x16GB HBM2 3200MT/s), BIOS Ver짯si짯on SE5C7411.86B.8424.D03.2208100444, ucode revision=0x2c000020, Cent짯OS Stream 8, Linux ver짯si짯on 5.19.0몉c6.0712.intel_땙ext.1.x86_64+server, YASK v3.05.07.

11 Up to 20% sys짯tem power savings uti짯li짯zing 4th Gen Xeon Sca짯lable with Opti짯mi짯zed Power mode on vs  off on sel짯ect workloads inclu짯ding Spec짯JBB, SPECINT and NIGNX key handshake.

12 AMD Milan: Tes짯ted by Numen짯ta as of 11/28/2022. 1몁ode, 2x AMD EPYC 7R13 on AWS m6a.48xlarge, 768 GB DDR4-3200, Ubun짯tu 20.04 Ker짯nel 5.15, Open짯VI짯NO 2022.3, BERT-Lar짯ge, Sequence Length 512, Batch Size 1

Intel짰 Xeon짰 8480+: Tes짯ted by Numen짯ta as of 11/28/2022. 1몁ode, 2x Intel짰 Xeon짰 8480+, 512 GB DDR5-4800, Ubun짯tu 22.04 Ker짯nel 5.17, Open짯VI짯NO 2022.3, Numen짯ta-Opti짯mi짯zed BERT-Lar짯ge, Sequence Length 512, Batch Size 1

Intel짰 Xeon짰 Max 9468: Tes짯ted by Numen짯ta as of 11/30/2022. 1몁ode, 2x Intel짰 Xeon짰 Max 9468, 128 GB HBM2e 3200 MT/s, Ubun짯tu 22.04 Ker짯nel 5.15, Open짯VI짯NO 2022.3, Numen짯ta-Opti짯mi짯zed BERT-Lar짯ge, Sequence Length 512, Batch Size 1

13 Intel짰 Xeon짰 8380: Test by Intel as of 10/28/2022. 1몁ode, 2x Intel짰 Xeon짰 8380 CPU, HT On, Tur짯bo On, Total Memo짯ry 256 GB (16x16GB 3200MT/s, Dual-Rank), BIOS Ver짯si짯on SE5C6200.86B.0020.P23.2103261309, ucode revision=0xd000270, Rocky Linux 8.6, Linux ver짯si짯on 4.18.0372.19.1.el8_6.crt1.x86_64

Intel짰 Xeon짰 CPU Max Series HBM: Test by Intel as of 10/28/2022. 1몁ode, 2x Intel짰 Xeon짰 Max 9480, HT On, Tur짯bo On, Total Memo짯ry 128 GB HBM2e, BIOS EGSDCRB1.DWR.0085.D12.2207281916, ucode 0xac000040, SUSE Linux Enter짯pri짯se Ser짯ver 15 SP3, Ker짯nel 5.3.18, oneA짯PI 2022.3.0

Intel짰 Data Cen짯ter GPU Max Series with DDR Host: Test by Intel as of 10/28/2022. 1몁ode, 2x Intel짰 Xeon짰 Max 9480, HT On, Tur짯bo On, Total Memo짯ry 1024 GB DDR5-4800 + 128 GB HBM2e, Memo짯ry Mode: Flat, HBM2e not used, 6x Intel짰 Data Cen짯ter GPU Max Series, BIOS EGSDCRB1.DWR.0085.D12.2207281916, ucode 0xac000040, Aga짯ma pvc-prq-54, SUSE Linux Enter짯pri짯se Ser짯ver 15 SP3, Ker짯nel 5.3.18, oneA짯PI 2022.3.0

Intel짰 Data Cen짯ter GPU Max Series with HBM Host: Test by Intel as of 10/28/2022. 1몁ode, 2x Intel짰 Xeon짰 Max 9480, HT On, Tur짯bo On, Total Memo짯ry 128 GB HBM2e, 6x Intel짰 Data Cen짯ter GPU Max Series, BIOS EGSDCRB1.DWR.0085.D12.2207281916, ucode 0xac000040, Aga짯ma pvc-prq-54, SUSE Linux Enter짯pri짯se Ser짯ver 15 SP3, Ker짯nel 5.3.18, oneA짯PI 2022.3.0