Announcing new AMD EPYC™-based Azure Virtual Machines

Micro­soft is com­mit­ted to giving our cus­to­mers indus­try-lea­ding per­for­mance for all their workloads. After being the first glo­bal cloud pro­vi­der to announ­ce the deploy­ment of AMD EPYC™ based Azu­re Vir­tu­al Machi­nes in 2017, we’ve been working tog­e­ther to con­ti­nue brin­ging the latest inno­va­ti­on to enter­pri­ses.

Today, we are announ­cing our second-genera­ti­on HB-seri­es Azu­re Vir­tu­al Machi­nes, HBv2, which fea­tures the latest AMD EPYC 7002 pro­ces­sor. Cus­to­mers will be able to incre­a­se HPC per­for­mance and sca­la­bi­li­ty to run mate­ri­al­ly lar­ger workloads on Azu­re. We’ll also be brin­ging the AMD 7002 pro­ces­sors and Rade­on Instinct GPUs to our fami­ly of cloud-based vir­tu­al desk­tops. Final­ly, our new Dav3 and Eav3-seri­es Azu­re Vir­tu­al Machi­nes, in pre­view today, pro­vi­de more cus­to­mer choice to meet a broad ran­ge of requi­re­ments for gene­ral pur­po­se workloads using the new AMD EPYC™ 7452 pro­ces­sor.

Our growing Azure HPC offerings

Cus­to­mers are choo­sing our Azu­re HPC offe­rings (HB-seri­es) incor­po­ra­ting first genera­ti­on AMD EPYC Nap­les for their per­for­mance and sca­la­bi­li­ty. We’ve seen a 33 per­cent memo­ry band­width advan­ta­ge with EPYC, and that’s a key fac­tor for many of our cus­to­mers’ HPC workloads. For examp­le, flu­id dyna­mics is one workload in which this advan­ta­ge is valu­able. Azu­re has an incre­a­sing num­ber of cus­to­mers for whom this is a core part of their R&D and even pro­duc­tion acti­vi­ties. On ANSYS Flu­ent, a wide­ly used flu­id dyna­mics app­li­ca­ti­on, we have mea­su­red EPYC-powe­red HB instan­ces deli­vering a 54x per­for­mance impro­ve­ment by sca­ling across near­ly 6,000 pro­ces­sor cores. And this is 24 per­cent fas­ter than a lea­ding bare-metal solu­ti­on with an iden­ti­cal Infi­ni­Band net­work. Addi­tio­nal­ly, ear­lier this year, Azu­re beca­me the first cloud to sca­le a tight­ly cou­pled HPC app­li­ca­ti­on to 10,000 cores. This is 10x hig­her than what had been pre­vious­ly pos­si­ble on any other cloud pro­vi­der. Azu­re cus­to­mers will be among the first to take advan­ta­ge of this capa­bi­li­ty to tack­le the toughest chal­len­ges and inno­va­te with pur­po­se.

New HPC, general purpose, and memory optimized Azure Virtual Machines

Azu­re is con­ti­nuing to incre­a­se its HPC capa­bi­li­ties, thanks in part to our col­la­bo­ra­ti­on with AMD. In preli­mi­na­ry bench­mar­king, HBv2 VMs fea­turing 120 CPUs from the second genera­ti­on EPYC pro­ces­sor are demons­tra­ting per­for­mance gains of over 100 per­cent on HPC workloads like flu­id dyna­mics and auto­mo­ti­ve crash test ana­ly­sis. HBv2 sca­la­bi­li­ty limits are also incre­a­sing with the cloud’s first deploy­ment of 200 Giga­bit Infi­ni­Band, thanks to the second genera­ti­on EPYC processor’s PCIe 4.0 capa­bi­li­ty. HBv2 vir­tu­al machi­nes (VMs) will sup­port up to 36,000 cores for MPI workloads in a sin­gle vir­tu­al machi­ne sca­le set, and up to 80,000 cores for our lar­gest cus­to­mers.

We’ll also be brin­ging AMD EPYC 7002 pro­ces­sor to our fami­ly of cloud-based remo­te desk­tops, pai­ring with the Rade­on MI25 GPU for cus­to­mers run­ning Win­dows-based envi­ron­ments. The new seri­es offers unpre­ce­den­ted GPU resour­cing fle­xi­bi­li­ty, giving cus­to­mers more choice than ever befo­re to size vir­tu­al machi­nes all the way from 1/8th of a sin­gle GPU up to a who­le GPU.

Final­ly, we are also announ­cing new Azu­re Vir­tu­al Machi­nes as part of the Dv3 and Ev3-series—optimized for gene­ral pur­po­se and memo­ry inten­si­ve workloads. The­se new VM sizes fea­ture AMD’s EPYC™ 7452 pro­ces­sor. The new gene­ral pur­po­se Da_v3 and Das_v3 Azu­re Vir­tu­al Machi­nes pro­vi­de up to 64 vCPUs, 256 GiBs of RAM, and 1,600 GiBs of SSD-based tem­pora­ry sto­rage. Addi­tio­nal­ly, the new memo­ry opti­mi­zed Ea_v3 and Eas_v3 Azu­re Vir­tu­al Machi­nes pro­vi­de up to 64 vCPUs, 432 GiBs of RAM, and 1,600 GiBs of SSD-based tem­pora­ry sto­rage. Both VM seri­es sup­port Pre­mi­um SSD disk sto­rage. The new VMs are cur­r­ent­ly in pre­view in the East US Azu­re regi­on and with avai­la­bi­li­ty com­ing soon to other regi­ons.

Da_v3 and Das_v3 vir­tu­al machi­nes can be used for a broad ran­ge of gene­ral-pur­po­se app­li­ca­ti­ons. Examp­le use cases inclu­de most enter­pri­se-gra­de app­li­ca­ti­ons, rela­tio­nal data­ba­ses, in-memo­ry caching, and ana­ly­tics. App­li­ca­ti­ons that demand fas­ter CPUs, bet­ter local disk per­for­mance or hig­her memo­ries can also bene­fit from the­se new VMs. Addi­tio­nal­ly, the Ea_v3 and Eas_v3 VM seri­es are opti­mi­zed for other lar­ge in-memo­ry busi­ness cri­ti­cal workloads.

Taking advantage of these new offerings