{"id":49445,"date":"2019-08-08T09:53:05","date_gmt":"2019-08-08T07:53:05","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=49445"},"modified":"2019-08-08T09:53:05","modified_gmt":"2019-08-08T07:53:05","slug":"announcing-new-amd-epyc-based-azure-virtual-machines","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/49445-announcing-new-amd-epyc-based-azure-virtual-machines\/","title":{"rendered":"Announcing new AMD<\/span> EPYC<\/span>\u2122-based Azure Virtual Machines"},"content":{"rendered":"

Micro\u00adsoft is com\u00admit\u00adted to giving our cus\u00adto\u00admers indus\u00adtry-lea\u00adding per\u00adfor\u00admance for all their workloads. After being the first glo\u00adbal cloud pro\u00advi\u00adder to announ\u00adce the deploy\u00adment of AMD<\/span> EPYC<\/span>\u2122 based Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes<\/a> in 2017, we\u2019ve been working tog\u00ade\u00adther to con\u00adti\u00adnue brin\u00adging the latest inno\u00adva\u00adti\u00adon to enterprises.<\/p>\n

Today, we are announ\u00adcing our second-gene\u00adra\u00adti\u00adon HB-series Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes, HBv2, which fea\u00adtures the latest AMD<\/span> EPYC<\/span> 7002 pro\u00adces\u00adsor<\/a>. Cus\u00adto\u00admers will be able to increase HPC<\/span> per\u00adfor\u00admance and sca\u00adla\u00adbi\u00adli\u00adty to run mate\u00adri\u00adal\u00adly lar\u00adger workloads on Azu\u00adre. We\u2019ll also be brin\u00adging the AMD<\/span> 7002 pro\u00adces\u00adsors and Rade\u00adon Instinct GPUs to our fami\u00adly of cloud-based vir\u00adtu\u00adal desk\u00adtops. Final\u00adly, our new Dav3 and Eav3-series Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes, in pre\u00adview today, pro\u00advi\u00adde more cus\u00adto\u00admer choice to meet a broad ran\u00adge of requi\u00adre\u00adments for gene\u00adral pur\u00adpo\u00adse workloads using the new AMD<\/span> EPYC<\/span>\u2122 7452 processor.<\/p>\n

Our growing Azure HPC<\/span> offerings<\/h2>\n

Cus\u00adto\u00admers are choo\u00adsing our Azu\u00adre HPC<\/span> offe\u00adrings (HB-series) incor\u00adpo\u00adra\u00adting first gene\u00adra\u00adti\u00adon AMD<\/span> EPYC<\/span> Nap\u00adles for their per\u00adfor\u00admance and sca\u00adla\u00adbi\u00adli\u00adty. We\u2019ve seen a 33 per\u00adcent memo\u00adry band\u00adwidth advan\u00adta\u00adge with EPYC<\/span>, and that\u2019s a key fac\u00adtor for many of our cus\u00adto\u00admers\u2019 HPC<\/span> workloads. For exam\u00adp\u00adle, flu\u00adid dyna\u00admics is one workload in which this advan\u00adta\u00adge is valuable. Azu\u00adre has an incre\u00adasing num\u00adber of cus\u00adto\u00admers for whom this is a core part of their R&<\/span>D and even pro\u00adduc\u00adtion acti\u00advi\u00adties. On ANSYS<\/span> Flu\u00adent, a wide\u00adly used flu\u00adid dyna\u00admics appli\u00adca\u00adti\u00adon, we have mea\u00adsu\u00adred EPYC-powered HB<\/span> ins\u00adtances deli\u00adve\u00adring a 54x per\u00adfor\u00admance impro\u00adve\u00adment by sca\u00adling across near\u00adly 6,000 pro\u00adces\u00adsor cores. And this is 24 per\u00adcent fas\u00adter than a lea\u00adding bare-metal solu\u00adti\u00adon with an iden\u00adti\u00adcal Infi\u00adni\u00adBand net\u00adwork. Addi\u00adtio\u00adnal\u00adly, ear\u00adlier this year, Azu\u00adre beca\u00adme the first cloud to sca\u00adle a tight\u00adly cou\u00adpled HPC<\/span> appli\u00adca\u00adti\u00adon to 10,000 cores. This is 10x hig\u00adher than what had been pre\u00advious\u00adly pos\u00adsi\u00adble on any other cloud pro\u00advi\u00adder. Azu\u00adre cus\u00adto\u00admers will be among the first to take advan\u00adta\u00adge of this capa\u00adbi\u00adli\u00adty to tack\u00adle the toug\u00adhest chal\u00adlenges and inno\u00adva\u00adte with purpose.<\/p>\n

New HPC<\/span>, general purpose, and memory optimized Azure Virtual Machines<\/h2>\n

Azu\u00adre is con\u00adti\u00adnuing to increase its HPC<\/span> capa\u00adbi\u00adli\u00adties, thanks in part to our col\u00adla\u00adbo\u00adra\u00adti\u00adon with AMD<\/span>. In preli\u00admi\u00adna\u00adry bench\u00admar\u00adking, HBv2 VMs fea\u00adturing 120 CPUs from the second gene\u00adra\u00adti\u00adon EPYC<\/span> pro\u00adces\u00adsor are demons\u00adt\u00adra\u00adting per\u00adfor\u00admance gains of over 100 per\u00adcent on HPC<\/span> workloads like flu\u00adid dyna\u00admics and auto\u00admo\u00adti\u00adve crash test ana\u00adly\u00adsis. HBv2 sca\u00adla\u00adbi\u00adli\u00adty limits are also incre\u00adasing with the cloud\u2019s first deploy\u00adment of 200 Giga\u00adbit Infi\u00adni\u00adBand, thanks to the second gene\u00adra\u00adti\u00adon EPYC<\/span> processor\u2019s PCIe 4.0 capa\u00adbi\u00adli\u00adty. HBv2 vir\u00adtu\u00adal machi\u00adnes (VMs) will sup\u00adport up to 36,000 cores for MPI<\/span> workloads in a sin\u00adgle vir\u00adtu\u00adal machi\u00adne sca\u00adle set, and up to 80,000 cores for our lar\u00adgest customers.<\/p>\n

We\u2019ll also be brin\u00adging AMD<\/span> EPYC<\/span> 7002 pro\u00adces\u00adsor to our fami\u00adly of cloud-based remo\u00adte desk\u00adtops, pai\u00adring with the Rade\u00adon MI25<\/span> GPU<\/span> for cus\u00adto\u00admers run\u00adning Win\u00addows-based envi\u00adron\u00adments. The new series offers unpre\u00adce\u00adden\u00adted GPU<\/span> resour\u00adcing fle\u00adxi\u00adbi\u00adli\u00adty, giving cus\u00adto\u00admers more choice than ever befo\u00adre to size vir\u00adtu\u00adal machi\u00adnes all the way from 1\/8th of a sin\u00adgle GPU<\/span> up to a who\u00adle GPU<\/span>.<\/p>\n

Final\u00adly, we are also announ\u00adcing new Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes as part of the Dv3 and Ev3-series\u2014optimized for gene\u00adral pur\u00adpo\u00adse and memo\u00adry inten\u00adsi\u00adve workloads. The\u00adse new VM<\/span> sizes fea\u00adture AMD<\/span>\u2019s EPYC<\/span>\u2122 7452 pro\u00adces\u00adsor. The new gene\u00adral pur\u00adpo\u00adse Da_v3 and Das_v3 Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes pro\u00advi\u00adde up to 64 vCPUs, 256 GiBs of RAM<\/span>, and 1,600 GiBs of SSD-based tem\u00adpo\u00adra\u00adry sto\u00adrage. Addi\u00adtio\u00adnal\u00adly, the new memo\u00adry opti\u00admi\u00adzed Ea_v3 and Eas_v3 Azu\u00adre Vir\u00adtu\u00adal Machi\u00adnes pro\u00advi\u00adde up to 64 vCPUs, 432 GiBs of RAM<\/span>, and 1,600 GiBs of SSD-based tem\u00adpo\u00adra\u00adry sto\u00adrage. Both VM<\/span> series sup\u00adport Pre\u00admi\u00adum SSD<\/span> disk sto\u00adrage. The new VMs are curr\u00adent\u00adly in pre\u00adview in the East US<\/span> Azu\u00adre regi\u00adon and with avai\u00adla\u00adbi\u00adli\u00adty coming soon to other regions.<\/p>\n

Da_v3 and Das_v3 vir\u00adtu\u00adal machi\u00adnes can be used for a broad ran\u00adge of gene\u00adral-pur\u00adpo\u00adse appli\u00adca\u00adti\u00adons. Exam\u00adp\u00adle use cases include most enter\u00adpri\u00adse-gra\u00adde appli\u00adca\u00adti\u00adons, rela\u00adtio\u00adnal data\u00adba\u00adses, in-memo\u00adry caching, and ana\u00adly\u00adtics. Appli\u00adca\u00adti\u00adons that demand fas\u00adter CPUs, bet\u00adter local disk per\u00adfor\u00admance or hig\u00adher memo\u00adries can also bene\u00adfit from the\u00adse new VMs. Addi\u00adtio\u00adnal\u00adly, the Ea_v3 and Eas_v3 VM<\/span> series are opti\u00admi\u00adzed for other lar\u00adge in-memo\u00adry busi\u00adness cri\u00adti\u00adcal workloads.<\/p>\n

Taking advantage of these new offerings<\/h2>\n