AMD Processors Accelerating Performance of Top Supercomputers Worldwide

Growing preference for EPYC꽓 processors resulted in the number of AMD-powered supercomputers growing 3.5x year-over-year

SANTA CLARA, Calif., Nov. 16, 2021 (GLOBE NEWSWIRE) During this year셲 Super짯com짯pu짯ting Con짯fe짯rence 2021 (SC21), AMD (NASDAQ: AMD) is show짯ca짯sing its expan짯ded pre짯sence and gro짯wing pre짯fe짯rence in the high per짯for짯mance com짯pu짯ting (HPC) indus짯try with the excep짯tio짯nal inno짯va짯ti짯on and adop짯ti짯on of AMD data cen짯ter pro짯ces짯sors and acce짯le짯ra짯tors. Cus짯to짯mers across the indus짯try con짯ti짯nue to expand their use of AMD EPYC꽓 pro짯ces짯sors and AMD Instinct꽓 acce짯le짯ra짯tors to power cut짯ting-edge rese짯arch nee짯ded to address some of the world셲 big짯gest chal짯lenges in cli짯ma짯te, life sci짯en짯ces, medi짯ci짯ne, and more.

Gro짯wing pre짯fe짯rence for AMD is show짯ca짯sed in the latest Top500 list. AMD now powers 73 super짯com짯pu짯ters, com짯pared to 21 in the Novem짯ber 2020 list, a more than 3x year-over-year increase. Addi짯tio짯nal짯ly, AMD powers four out of the top ten most powerful super짯com짯pu짯ters in the world, as well as the most powerful super짯com짯pu짯ter in EMEA. Final짯ly, AMD EPYC 7003 series pro짯ces짯sors, which laun짯ched eight months ago, are uti짯li짯zed by 17 of the 75 AMD powered super짯com짯pu짯ters in the list, demons짯t짯ra짯ting the rapid adop짯ti짯on of the latest gene짯ra짯ti짯on of EPYC processors.

The demands of super짯com짯pu짯ting users have increased expo짯nen짯ti짯al짯ly as the world seeks to acce짯le짯ra짯te rese짯arch, redu짯cing the time to dis짯co짯very of valuable infor짯ma짯ti짯on, said For짯rest Nor짯rod, seni짯or vice pre짯si짯dent and gene짯ral mana짯ger, Data Cen짯ter and Embedded Solu짯ti짯ons Busi짯ness Group, AMD. 쏻ith AMD EPYC CPUs and Instinct acce짯le짯ra짯tors, we con짯ti짯nue to evol짯ve our pro짯duct offe짯ring to push the boun짯da짯ries of data cen짯ter tech짯no짯lo짯gies enab짯ling fas짯ter rese짯arch, bet짯ter out짯co짯mes and more impact on the world.

AMD has also been reco짯gni짯zed in the annu짯al HPC짯wire Rea짯ders and Edi짯tors Choice Awards at SC21. The com짯pa짯ny won ten awards inclu짯ding Best Sus짯taina짯bi짯li짯ty Inno짯va짯ti짯on in HPC, Best HPC Ser짯ver Pro짯duct and the Out짯stan짯ding Lea짯der짯ship in HPC award pre짯sen짯ted to Pre짯si짯dent and CEO Dr. Lisa Su.

Expan짯ding Cus짯to짯mer Base
AMD is enga짯ged broad짯ly across the HPC indus짯try to deli짯ver the per짯for짯mance and effi짯ci짯en짯cy of AMD EPYC and AMD Instinct pro짯ducts, along with the ROCm꽓 open eco짯sys짯tem, to speed rese짯arch. Through high-pro짯fi짯le instal짯la짯ti짯ons like the ongo짯ing deploy짯ment of Oak Ridge Natio짯nal Laboratory셲 쏤ron짯tier super짯com짯pu짯ter, AMD is brin짯ging the com짯pu짯te tech짯no짯lo짯gies and per짯for짯mance nee짯ded to sup짯port deve짯lo짯p짯ments in cur짯rent and future rese짯arch across the world. High짯lights of 쏤ron짯tier and other new HPC sys짯tems in the indus짯try include:

A Year of Breakth짯rough Pro짯ducts and Research 
This year AMD laun짯ched its AMD EPYC 7003 series pro짯ces짯sor, the world셲 hig짯hest-per짯forming ser짯ver pro짯ces짯sor.1 Sin짯ce then, the짯re has been over짯whel짯ming adop짯ti짯on from part짯ners across the indus짯try who are dri짯ving dis짯co짯veries in bio짯me짯di짯ci짯ne, pre짯dic짯ting natu짯ral dis짯as짯ters, clean ener짯gy, semi짯con짯duc짯tors, microelec짯tro짯nics and more.

Expan짯ding on the fea짯tures of the EPYC 7003 series pro짯ces짯sor, AMD recent짯ly pre짯view짯ed the 3rd Gen EPYC pro짯ces짯sor with AMD 3D V멵ache. By uti짯li짯zing inno짯va짯ti짯ve pack짯a짯ging tech짯no짯lo짯gy, which lay짯ers the L3 cache in EPYC 7003 series pro짯ces짯sors, AMD 3D V멌ache tech짯no짯lo짯gy offers enhan짯ced per짯for짯mance for the tech짯ni짯cal com짯pu짯ting workloads pre짯va짯lent in HPC. Micro짯soft Azu짯re HPC vir짯tu짯al machi짯nes fea짯turing 3rd Gen EPYC with AMD 3D V멌ache are curr짯ent짯ly available in Pri짯va짯te Pre짯view and will be available glo짯bal짯ly soon.

AMD also unvei짯led the world셲 fas짯test HPC and AI acce짯le짯ra짯tor2, AMD Instinct MI250X. Desi짯gned with the AMD CDNA꽓 2 archi짯tec짯tu짯re, the AMD Instinct MI200 series acce짯le짯ra짯tors deli짯ver up to 4.9x the peak FP64 per짯for짯mance ver짯sus com짯pe짯ti짯ti짯ve data cen짯ter acce짯le짯ra짯tors, which is cri짯ti짯cal for HPC appli짯ca짯ti짯ons like wea짯ther mode짯ling2. The AMD Instinct MI200 series acce짯le짯ra짯tors are also the first to have over 100GB high-band짯width memo짯ry capa짯ci짯ty, deli짯ve짯ring up to 3.2 tera짯bytes per second, the industry셲 best aggre짯ga짯te band짯width3.

Sup짯port짯ing Resources

About AMD
For more than 50 years AMD has dri짯ven inno짯va짯ti짯on in high-per짯for짯mance com짯pu짯ting, gra짯phics and visua짯liza짯ti짯on tech짯no짯lo짯gies the buil짯ding blocks for gam짯ing, immersi짯ve plat짯forms and the dat짯a짯cen짯ter. Hundreds of mil짯li짯ons of con짯su짯mers, lea짯ding For짯tu짯ne 500 busi짯nesses and cut짯ting-edge sci짯en짯ti짯fic rese짯arch faci짯li짯ties around the world rely on AMD tech짯no짯lo짯gy dai짯ly to impro짯ve how they live, work and play. AMD employees around the world are focu짯sed on buil짯ding gre짯at pro짯ducts that push the boun짯da짯ries of what is pos짯si짯ble. For more infor짯ma짯ti짯on about how AMD is enab짯ling today and inspi짯ring tomor짯row, visit the AMD (NASDAQ: AMD) web짯site, blog, Face짯book and Twit짯ter pages.

AMD, the AMD Arrow logo, AMD CDNA, EPYC, AMD Instinct and com짯bi짯na짯ti짯ons the짯reof are trade짯marks of Advan짯ced Micro Devices, Inc. Other names are for infor짯ma짯tio짯nal pur짯po짯ses only and may be trade짯marks of their respec짯ti짯ve owners.

_____________________________

1 MLN-016: Results as of 01/28/2021 using SPECrate짰2017_int_base. The AMD EPYC 7763 a mea짯su짯red esti짯ma짯ted score of 798 is hig짯her than the cur짯rent hig짯hest 2P ser짯ver with an AMD EPYC 7H12 and a score of 717, https://spec.org/cpu2017/results/res2020q2/cpu2017-2020052522554.pdf. OEM published score(s) for 3rd Gen EPYC may vary. SPEC짰, SPE짯Cra짯te짰 and SPEC CPU짰 are regis짯tered trade짯marks of the Stan짯dard Per짯for짯mance Eva짯lua짯ti짯on Cor짯po짯ra짯ti짯on. See www.spec.org for more information.
2 MI200-01: World셲 fas짯test data cen짯ter GPU is the AMD Instinct꽓 MI250X. Cal짯cu짯la짯ti짯ons con짯duc짯ted by AMD Per짯for짯mance Labs as of Sep 15, 2021, for the AMD Instinct꽓 MI250X (128GB HBM2e OAM modu짯le) acce짯le짯ra짯tor at 1,700 MHz peak boost engi짯ne clock resul짯ted in 95.7 TFLOPS peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on (FP64 Matrix), 47.9 TFLOPS peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on (FP64), 95.7 TFLOPS peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on matrix (FP32 Matrix), 47.9 TFLOPS peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on (FP32), 383.0 TFLOPS peak theo짯re짯ti짯cal half pre짯cis짯i짯on (FP16), and 383.0 TFLOPS peak theo짯re짯ti짯cal Bfloat16 for짯mat pre짯cis짯i짯on (BF16) floa짯ting-point per짯for짯mance. Cal짯cu짯la짯ti짯ons con짯duc짯ted by AMD Per짯for짯mance Labs as of Sep 18, 2020 for the AMD Instinct꽓 MI100 (32GB HBM2 PCIe짰 card) acce짯le짯ra짯tor at 1,502 MHz peak boost engi짯ne clock resul짯ted in 11.54 TFLOPS peak theo짯re짯ti짯cal dou짯ble pre짯cis짯i짯on (FP64), 46.1 TFLOPS peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on matrix (FP32), 23.1 TFLOPS peak theo짯re짯ti짯cal sin짯gle pre짯cis짯i짯on (FP32), 184.6 TFLOPS peak theo짯re짯ti짯cal half pre짯cis짯i짯on (FP16) floa짯ting-point per짯for짯mance. Published results on the NVi짯dia Ampere A100 (80GB) GPU acce짯le짯ra짯tor, boost engi짯ne clock of 1410 MHz, resul짯ted in 19.5 TFLOPS peak dou짯ble pre짯cis짯i짯on ten짯sor cores (FP64 Ten짯sor Core), 9.7 TFLOPS peak dou짯ble pre짯cis짯i짯on (FP64). 19.5 TFLOPS peak sin짯gle pre짯cis짯i짯on (FP32), 78 TFLOPS peak half pre짯cis짯i짯on (FP16), 312 TFLOPS peak half pre짯cis짯i짯on (FP16 Ten짯sor Flow), 39 TFLOPS peak Bfloat 16 (BF16), 312 TFLOPS peak Bfloat16 for짯mat pre짯cis짯i짯on (BF16 Ten짯sor Flow), theo짯re짯ti짯cal floa짯ting-point per짯for짯mance. The TF32 data for짯mat is not IEEE com짯pli짯ant and not included in this com짯pa짯ri짯son. https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/nvidia-ampere-architecture-whitepaper.pdf, page 15, Table 1.
3 MI200-07: Cal짯cu짯la짯ti짯ons con짯duc짯ted by AMD Per짯for짯mance Labs as of Sep 21, 2021, for the AMD Instinct꽓 MI250X and MI250 (128GB HBM2e) OAM acce짯le짯ra짯tors desi짯gned with AMD CDNA꽓 2 6nm Fin짯Fet pro짯cess tech짯no짯lo짯gy at 1,600 MHz peak memo짯ry clock resul짯ted in 128GB HBM2e memo짯ry capa짯ci짯ty and 3.2768 TFLOPS peak theo짯re짯ti짯cal memo짯ry band짯width per짯for짯mance. MI250/MI250X memo짯ry bus inter짯face is 4,096 bits times 2 die and memo짯ry data rate is 3.20 Gbps for total memo짯ry band짯width of 3.2768 TB/s ((3.20 Gbps*(4,096 bits*2))/8).The hig짯hest published results on the NVi짯dia Ampere A100 (80GB) SXM GPU acce짯le짯ra짯tor resul짯ted in 80GB HBM2e memo짯ry capa짯ci짯ty and 2.039 TB/s GPU memo짯ry band짯width per짯for짯mance. https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/a100/pdf/nvidia-a100-datasheet-us-nvidia-1758950-r4-web.pdf