{"id":65376,"date":"2022-05-26T15:04:51","date_gmt":"2022-05-26T13:04:51","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=65376"},"modified":"2022-05-26T15:04:51","modified_gmt":"2022-05-26T13:04:51","slug":"amd-instinct-mi200-adopted-for-large-scale-ai-training-in-microsoft-azure","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/65376-amd-instinct-mi200-adopted-for-large-scale-ai-training-in-microsoft-azure\/","title":{"rendered":"AMD<\/span> Instinct\u2122 MI200<\/span> Adopted for Large-Scale AI<\/span> Training in Microsoft Azure"},"content":{"rendered":"

\ua7f7 Micro\u00adsoft Azu\u00adre is the first public cloud to deploy AMD<\/span> Instinct MI200<\/span> acce\u00adle\u00adra\u00adtors \ua7f7<\/em><\/p>\n

SANTA<\/span> CLARA<\/span>, Calif., May 26, 2022 (GLOBE<\/span> NEWSWIRE<\/span>) \u2014 AMD<\/span><\/a> (NASDAQ<\/span>: AMD<\/span>) and Micro\u00adsoft con\u00adtin\u00adued their col\u00adla\u00adbo\u00adra\u00adti\u00adon in the cloud, with Micro\u00adsoft announ\u00adcing the use<\/a> of AMD<\/span> Instinct\u2122 MI200<\/span> acce\u00adle\u00adra\u00adtors to power lar\u00adge sca\u00adle AI<\/span> trai\u00adning workloads. In addi\u00adti\u00adon, Micro\u00adsoft announ\u00adced it is working clo\u00adse\u00adly with the PyTorch Core team and AMD<\/span> data cen\u00adter soft\u00adware team to opti\u00admi\u00adze the per\u00adfor\u00admance and deve\u00adlo\u00adper expe\u00adri\u00adence for cus\u00adto\u00admers run\u00adning PyTorch on Micro\u00adsoft Azu\u00adre and ensu\u00adre that deve\u00adlo\u00adpers\u2019 PyTorch pro\u00adjects take advan\u00adta\u00adge of the per\u00adfor\u00admance and fea\u00adtures of AMD<\/span> Instinct accelerators.<\/p>\n

\u201c<\/span>AMD<\/span> Instinct MI200<\/span> acce\u00adle\u00adra\u00adtors pro\u00advi\u00adde cus\u00adto\u00admers with cut\u00adting-edge AI<\/span> and HPC<\/span> per\u00adfor\u00admance and will power many of the world\u2019s fas\u00adtest super\u00adcom\u00adpu\u00adter sys\u00adtems that push the boun\u00adda\u00adries of sci\u00adence,\u201d said Brad McCre\u00addie, cor\u00adpo\u00adra\u00adte vice pre\u00adsi\u00addent, Data Cen\u00adter and Acce\u00adle\u00adra\u00adted Pro\u00adces\u00adsing, AMD<\/span>. \u201cOur work with Micro\u00adsoft to enable lar\u00adge-sca\u00adle AI<\/span> trai\u00adning and infe\u00adrence in the cloud not only high\u00adlights the expan\u00adsi\u00adve tech\u00adno\u00adlo\u00adgy col\u00adla\u00adbo\u00adra\u00adti\u00adon bet\u00adween the two com\u00adpa\u00adnies, but the per\u00adfor\u00admance capa\u00adbi\u00adli\u00adties of AMD<\/span> Instinct MI200<\/span> acce\u00adle\u00adra\u00adtors and how they will ulti\u00adm\u00adate\u00adly help cus\u00adto\u00admers advan\u00adce the gro\u00adwing demands for AI<\/span> workloads.\u201d<\/p>\n

\u201c<\/span>We\u2019re proud to build upon our long-term com\u00admit\u00adment to inno\u00adva\u00adti\u00adon with AMD<\/span> and make Azu\u00adre the first public cloud to deploy clus\u00adters of the AMD<\/span> Instinct MI200<\/span> acce\u00adle\u00adra\u00adtor for lar\u00adge sca\u00adle AI<\/span> trai\u00adning,\u201d said Eric Boyd, cor\u00adpo\u00adra\u00adte vice pre\u00adsi\u00addent, Azu\u00adre AI<\/span>, Micro\u00adsoft. \u201cWe have star\u00adted test\u00ading the MI200<\/span> with our own AI<\/span> workloads and are see\u00ading gre\u00adat per\u00adfor\u00admance, and we look for\u00adward to con\u00adti\u00adnuing our col\u00adla\u00adbo\u00adra\u00adti\u00adon with AMD<\/span> to bring cus\u00adto\u00admers more per\u00adfor\u00admance, choice and fle\u00adxi\u00adbi\u00adli\u00adty for their AI<\/span> needs.\u201d<\/p>\n

The AMD<\/span> Instinct MI200<\/span> acce\u00adle\u00adra\u00adtors join other AMD<\/span> pro\u00adducts being used at Micro\u00adsoft Azu\u00adre inclu\u00adding the new\u00adly announ\u00adced Azu\u00adre HBv3 vir\u00adtu\u00adal machi\u00adnes<\/a> which use 3rd <\/sup>Gen AMD<\/span> EPYC<\/span>\u2122 pro\u00adces\u00adsors with AMD<\/span> 3D<\/span> V\u2011Cache\u2122 Tech\u00adno\u00adlo\u00adgy, and num\u00ade\u00adrous other ins\u00adtances<\/a> for con\u00adfi\u00adden\u00adti\u00adal com\u00adpu\u00adting, gene\u00adral pur\u00adpo\u00adse workloads, memo\u00adry bound workloads, visu\u00adal workloads and more.<\/p>\n

Sup\u00adport\u00ading Resources<\/strong><\/p>\n