{"id":59828,"date":"2020-11-16T15:02:21","date_gmt":"2020-11-16T14:02:21","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=59828"},"modified":"2020-11-16T21:53:36","modified_gmt":"2020-11-16T20:53:36","slug":"amd-epyc-processors-and-new-amd-instinct-mi100-accelerator-redefine-performance-for-hpc-and-scientific-research","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/59828-amd-epyc-processors-and-new-amd-instinct-mi100-accelerator-redefine-performance-for-hpc-and-scientific-research\/","title":{"rendered":"AMD<\/span> EPYC<\/span>\u2122 Processors and New AMD<\/span> Instinct\u2122 MI100<\/span> Accelerator Redefine Performance for HPC<\/span> and Scientific Research"},"content":{"rendered":"

\u2014Sup\u00adport for next gene\u00adra\u00adti\u00adon AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors, code\u00adna\u00admed \u2018Milan,\u2019 in new HBv3 Vir\u00adtu\u00adal Machi\u00adnes announ\u00adced by Microsoft\u2014<\/i><\/span><\/span><\/span><\/p>\n

\u2014AMD<\/span> Instinct\u2122 MI100<\/span> acce\u00adle\u00adra\u00adtor is first acce\u00adle\u00adra\u00adtor to use new AMD<\/span> CDNA<\/span> archi\u00adtec\u00adtu\u00adre dedi\u00adca\u00adted to high-per\u00adfor\u00admance com\u00adpu\u00adting (HPC<\/span>) workloads\u2014<\/i><\/span><\/span><\/span><\/p>\n

SANTA<\/span> CLARA<\/span>, Calif. <\/b><\/span><\/span>\u2014 Novem\u00adber 16, 2020 \u2014 <\/b><\/span>During this year\u2019s SC20<\/span> vir\u00adtu\u00adal trade\u00adshow, <\/span>AMD<\/span><\/a><\/u><\/span><\/span> (NASDAQ<\/span>: AMD<\/span>) is show\u00adca\u00adsing its lea\u00adder\u00adship in the high per\u00adfor\u00admance com\u00adpu\u00adting (HPC<\/span>) indus\u00adtry. It laun\u00adched the new AMD<\/span> Instinct\u2122 MI100<\/span> acce\u00adle\u00adra\u00adtor with ROCm\u2122 4.0 open eco\u00adsys\u00adtem sup\u00adport, show\u00adca\u00adsed a gro\u00adwing list of AMD<\/span> EPYC<\/span>\u2122 CPU<\/span> and AMD<\/span> Instinct acce\u00adle\u00adra\u00adtor based deploy\u00adments, and high\u00adligh\u00adted the growth AMD<\/span> EPYC<\/span> CPUs and AMD<\/span> Instinct\u2122 GPUs on the TOP500<\/span> list. <\/span>AMD<\/span> also remains on track to begin volu\u00adme ship\u00adments of the 3<\/span>rd<\/span><\/sup> Gen EPYC<\/span> pro\u00adces\u00adsors with \u201cZen 3\u201d core to sel\u00adect HPC<\/span> and cloud cus\u00adto\u00admers this quar\u00adter in advan\u00adce of the expec\u00adted public launch in Q1<\/span> 2021, ali\u00adgned with OEM<\/span> availability.<\/span><\/span><\/span><\/p>\n

The new <\/span>AMD<\/span> Instinct\u2122 MI100<\/span> acce\u00adle\u00adra\u00adtor<\/span><\/a><\/u><\/span>, is the world\u2019s fas\u00adtest HPC<\/span> GPU<\/span> acce\u00adle\u00adra\u00adtor for sci\u00aden\u00adti\u00adfic workloads<\/span> and the first to sur\u00adpass the 10 tera\u00adflops (FP64<\/span>) per\u00adfor\u00admance bar\u00adri\u00ader<\/span> i<\/sup><\/a>. Built on the new AMD<\/span> <\/span>CDNA<\/span> archi\u00adtec\u00adtu\u00adre<\/a>,<\/span><\/u><\/span> the AMD<\/span> Instinct MI100<\/span> GPU<\/span> enables a new class of acce\u00adle\u00adra\u00adted sys\u00adtems for HPC<\/span> and AI<\/span> when pai\u00adred with 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors. Sup\u00adport\u00aded by new acce\u00adle\u00adra\u00adted com\u00adpu\u00adte plat\u00adforms from Dell, HPE<\/span>, Giga\u00adbyte and Super\u00admi\u00adcro, the MI100<\/span>, com\u00adbi\u00adned with AMD<\/span> EPYC<\/span> CPUs and ROCm 4.0 soft\u00adware, is desi\u00adgned to pro\u00adpel new dis\u00adco\u00adveries ahead of the exas\u00adca\u00adle era.<\/span><\/span><\/span><\/p>\n

\u201c<\/span>No two cus\u00adto\u00admers are the same in HPC<\/span>, and AMD<\/span> is pro\u00advi\u00adding a path to today\u2019s most advan\u00adced tech\u00adno\u00adlo\u00adgies and capa\u00adbi\u00adli\u00adties that are cri\u00adti\u00adcal to sup\u00adport their HPC<\/span> work, from small clus\u00adters on pre\u00admi\u00adse, to vir\u00adtu\u00adal machi\u00adnes in the cloud, all the way to exas\u00adca\u00adle super\u00adcom\u00adpu\u00adters,\u201d said For\u00adrest Nor\u00adrod, seni\u00ador vice pre\u00adsi\u00addent and gene\u00adral mana\u00adger, Data Cen\u00adter and Embedded Solu\u00adti\u00adons Busi\u00adness Group, AMD<\/span>. \u201cCom\u00adbi\u00adning AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors and Instinct acce\u00adle\u00adra\u00adtors <\/span><\/span><\/span>with cri\u00adti\u00adcal appli\u00adca\u00adti\u00adon soft\u00adware and deve\u00adlo\u00adp\u00adment tools enables AMD<\/span> to deli\u00adver lea\u00adder\u00adship per\u00adfor\u00admance for HPC<\/span> workloads.\u201d<\/span><\/span><\/p>\n

\n

AMD<\/span> and Micro\u00adsoft Azu\u00adre Power HPC<\/span> In the Cloud<\/b><\/span><\/span><\/span><\/p>\n

Azu\u00adre is using 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors to power its <\/span>HBv2 vir\u00adtu\u00adal machi\u00adnes (VMs) for HPC<\/span> workloads<\/span><\/a><\/u><\/span>.<\/span><\/u> The\u00adse VMs<\/span> offer up to 2x the per\u00adfor\u00admance of first-gene\u00adra\u00adti\u00adon HB-series vir\u00adtu\u00adal machi\u00adnesi<\/sup><\/a>, can sup\u00adport up to 80,000 cores for MPI<\/span> jobsii<\/sup><\/a>, and take advan\u00adta\u00adge of 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors\u2019<\/span> up to 45% more memo\u00adry band\u00adwidth than com\u00adpa\u00adra\u00adble x86 alter\u00adna\u00adti\u00advesiii<\/sup><\/a>.<\/span><\/span><\/span><\/p>\n

HBv2 VMs are used by num\u00ade\u00adrous cus\u00adto\u00admers inclu\u00adding The Uni\u00adver\u00adsi\u00adty of Illi\u00adnois at Urbana-Champaign\u2019s Beck\u00adman Insti\u00adtu\u00adte for Advan\u00adced Sci\u00adence &<\/span> Tech\u00adno\u00adlo\u00adgy which <\/span>used 86,400 cores to model a plant virus that pre\u00advious\u00adly requi\u00adred a lea\u00adder\u00adship class super\u00adcom\u00adpu\u00adter<\/span><\/a><\/u><\/span> and the <\/span>U.S. Navy which rapidly deploys and sca\u00adles enhan\u00adced wea\u00adther and oce\u00adan pat\u00adtern pre\u00addic\u00adtions on demand<\/a>.<\/span><\/u><\/span> HBv2 powered by 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors also pro\u00advi\u00addes the bulk of the CPU<\/span> com\u00adpu\u00adte power for the<\/span>Ope\u00adnAI envi\u00adron\u00adment Micro\u00adsoft announ\u00adced<\/a><\/span><\/u><\/span> ear\u00adlier this year. <\/span><\/span><\/span><\/p>\n

AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors have also <\/span>hel\u00adped HBv2 reach new cloud HPC<\/span> mile\u00adsto\u00adnes<\/span><\/a><\/u><\/span>, such as a new record for Cloud MPI<\/span> sca\u00adling results with NAMD<\/span>, Top 20 results on the Graph500, and the first 1 terabyte\/sec cloud HPC<\/span> par\u00adal\u00adlel file\u00adsys\u00adtem. Across the\u00adse and other appli\u00adca\u00adti\u00adon bench\u00admarks, HBv2 is deli\u00adve\u00adring 12x hig\u00adher sca\u00adling than found else\u00adwhe\u00adre on the public cloud.<\/span><\/span><\/span><\/p>\n

Adding on to its exis\u00adting HBv2 HPC<\/span> vir\u00adtu\u00adal machi\u00adne powered by 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors, Azu\u00adre announ\u00adced it will uti\u00adli\u00adze next gene\u00adra\u00adti\u00adon AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors, code\u00adna\u00admed \u2018Milan\u2019, for future HB-series VM<\/span> pro\u00adducts for HPC<\/span>. <\/span><\/span><\/span><\/p>\n

You can see more about the <\/span>AMD<\/span> and Azu\u00adre col\u00adla\u00adbo\u00adra\u00adti\u00adon in this video<\/span><\/a><\/u><\/span> with Jason Zan\u00adder of Azu\u00adre and Lisa Su of AMD<\/span>. <\/span><\/span><\/span><\/p>\n

AMD<\/span> Con\u00adti\u00adnues to Be the Choice for HPC<\/span> <\/b><\/span><\/span><\/span><\/p>\n

AMD<\/span> EPYC<\/span> pro\u00adces\u00adsors and Instinct acce\u00adle\u00adra\u00adtors have the per\u00adfor\u00admance and capa\u00adbi\u00adli\u00adties to sup\u00adport num\u00ade\u00adrous HPC<\/span> workloads across a varie\u00adty of imple\u00admen\u00adta\u00adti\u00adons. From small clus\u00adters at rese\u00adarch cen\u00adters, to com\u00admer\u00adcial HPC<\/span>, to off pre\u00admi\u00adse and in the cloud, to exas\u00adca\u00adle com\u00adpu\u00adting, AMD<\/span> con\u00adti\u00adnues to pro\u00advi\u00adde per\u00adfor\u00admance and choice for HPC<\/span> solutions. <\/span><\/span><\/span><\/p>\n

Hew\u00adlett Packard Enter\u00adpri\u00adse (HPE<\/span>), CSC<\/span> Fin\u00adland and EuroHPC <\/span>recent\u00adly intro\u00addu\u00adced a new pre-exas\u00adca\u00adle sys\u00adtem, LUMI<\/span><\/a>.<\/span><\/u><\/span> Based on the HPE<\/span> Cray EX<\/span> super\u00adcom\u00adpu\u00adter archi\u00adtec\u00adtu\u00adre, LUMI<\/span> will use next gene\u00adra\u00adti\u00adon AMD<\/span> EPYC<\/span> CPUs and Instinct acce\u00adle\u00adra\u00adtors and is expec\u00adted to pro\u00advi\u00adde a <\/span><\/span><\/span>peak per\u00adfor\u00admance of 552 peta\u00adflops when it comes online in 2021, making it one of the fas\u00adtest super\u00adcom\u00adpu\u00adters in the world.<\/span><\/span><\/p>\n

Bey\u00adond LUMI<\/span>, AMD<\/span> powered HPC<\/span> sys\u00adtems con\u00adti\u00adnue to grow in volu\u00adme. Sin\u00adce SC19<\/span>, the\u00adre have been more than 15 super\u00adcom\u00adpu\u00adting sys\u00adtems announ\u00adced using AMD<\/span> EPYC<\/span> CPUs, Instinct GPUs, or both. A high\u00adlight of the sys\u00adtems includes <\/span><\/span><\/span><\/p>\n