In die­ser Sek­ti­on wer­den Pres­se­mit­tei­lun­gen der Her­stel­ler 1:1 ver­öf­fent­licht. Pla­net 3DNow! über­nimmt weder für die Rich­tig­keit der Anga­ben noch für die Art der For­mu­lie­rung jed­we­de Ver­ant­wor­tung. Die Anga­ben kön­nen sub­jek­tiv sein und geben nicht not­wen­di­ger­wei­se die Mei­nung der auf Pla­net 3DNow! schrei­ben­den Autoren wieder.

AMD FirePro GPUs Enable GSI Research Facility’s Heavy Ion Studies

Verfasst vonDr@

AMD Fire­Pro S9150 GPU powers more than 3 peta­FLOPS super­com­pu­ter clus­ter for Open­CL™ appli­ca­ti­ons at GSI

SUNNYVALE, Calif. 11/13/2014

AMD (NYSE: AMD) lea­der­ship in hete­ro­ge­neous high per­for­mance com­pu­ting (HPC) is deli­ve­ring more than 3 peta­FLOPS of AMD Fire­Pro™ S9150 ser­ver GPU per­for­mance for the GSI Helm­holtz­zen­trum für Schwer­io­nen­for­schung GmbH in sup­port of phy­sics rese­arch rela­ted to hea­vy ion stu­dies. Each of the 160 ASUS ESC4000 G2S nodes1 of the new L‑CSC clus­ter include four AMD Fire­Pro S9150 ser­ver GPUs, for a total GPU peak of 3.25 peta­FLOPS sin­gle pre­cis­i­on and 1.62 peta­FLOPS dou­ble pre­cis­i­on per­for­mance. The clus­ter is enab­ling Lat­ti­ce Quan­tum Chro­mo Dyna­mics (Lat­ti­ce QCD) com­pu­ta­tio­nal rese­arch using one of the fas­test Open­CL™ imple­men­ta­ti­ons for rese­arch appli­ca­ti­ons in the world. Lat­ti­ce QCD cal­cu­la­ti­ons are appli­ca­ble to a ran­ge of stu­dies in high ener­gy and hea­vy ion physics.

“AMD is proud to col­la­bo­ra­te with ASUS, the Frank­furt Insti­tu­te for Advan­ced Stu­dies, (FIAS) and GSI to sup­port such important phy­sics and com­pu­ter sci­ence rese­arch,” said David Cum­mings, seni­or direc­tor and gene­ral mana­ger, pro­fes­sio­nal gra­phics, AMD. “This instal­la­ti­on reaf­firms AMD’s lea­ding role in HPC with the imple­men­ta­ti­on of the AMD Fire­Pro S9150 ser­ver GPUs in this three peta­FLOPS super­com­pu­ter clus­ter. AMD and ASUS are enab­ling Open­CL™ appli­ca­ti­ons for cri­ti­cal sci­ence rese­arch usa­ge for this clus­ter. We’re com­mit­ted to buil­ding our HPC lea­der­ship posi­ti­on in the indus­try as a fore­most pro­vi­der of com­pu­ting appli­ca­ti­ons, tools and technologies.”

“We had excel­lent coope­ra­ti­on with ASUS and AMD to make this pro­ject hap­pen in such a short time­frame,” said Pro­fes­sor Doc­tor Vol­ker Lin­denstruth, pro­fes­sor at Goe­the Uni­ver­si­ty of Frank­furt and chair­man of Frank­furt Insti­tu­te for Advan­ced Stu­dies. “The ASUS ESC4000 G2S ser­vers and the AMD Fire­Pro S9150 GPUs are an extre­me­ly powerful basis for the L‑CSC clus­ter and they pro­vi­de the com­pu­te capa­bi­li­ties we need for our rese­arch. The lar­ge 16GB of memo­ry of the AMD Fire­Pro S9150 ser­ver GPUs allows us to run most LQCD com­pu­ta­ti­ons on one GPU wit­hout inter-GPU and inter-node com­mu­ni­ca­ti­on resul­ting in very effi­ci­ent LQCD appli­ca­ti­on ope­ra­ti­on. We have cho­sen a mul­ti-GPU approach with four GPU boards per ser­ver to enable an extre­me­ly cost-effi­ci­ent and ener­gy-effi­ci­ent configuration.”

“ASUS is pas­sio­na­te about the HPC field, in the same way that AMD is dedi­ca­ted to deli­ve­ring extre­me-per­for­mance GPU solu­ti­ons. Tog­e­ther, ASUS and AMD have rea­ched a new level with the GSI pro­ject, reve­al­ing the true power of ASUS ESC4000 G2S GPU ser­vers and AMD Fire­Pro S9150 GPUs,” said Tom Lin, ASUS Gene­ral Mana­ger. “We’re proud to offer our cus­to­mers solu­ti­ons that deli­ver unpar­al­le­led per­for­mance and utmost ener­gy effi­ci­en­cy. The­re is no doubt that our joint GSI pro­ject deli­vers top-tier per­for­mance for cri­ti­cal HPC applications.”

GSI (Gesell­schaft für Schwer­io­nen­for­schung), loca­ted in Darm­stadt, Ger­ma­ny, is a rese­arch faci­li­ty for hea­vy ion rese­arch inclu­ding can­cer the­ra­py. The com­pu­te clus­ter was desi­gned and built in con­junc­tion with FIAS, a rese­arch insti­tu­te for phy­sics, neu­ro sci­ence, life sci­ence, and com­pu­ter sci­ence. Within the next few years, GSI will build FAIR (Faci­li­ty for Anti-Pro­ton and Hea­vy-Ion Rese­arch), a new par­tic­le acce­le­ra­tor and detec­tor faci­li­ty. Rese­ar­chers from around the world use this faci­li­ty for expe­ri­ments that help them make fasci­na­ting rese­arch discoveries.

High ener­gy phy­sics expe­ri­ments, such as at GSI, need huge com­pu­te farms for simu­la­ti­ons. The L‑CSC clus­ter is aimed at QCD com­pu­ta­ti­ons, which requi­re extre­me­ly high memo­ry band­width. The imple­men­ta­ti­on of the Lat­ti­ce QCD algo­rithm was deve­lo­ped at FIAS using an Open­CL™ based imple­men­ta­ti­on for por­ta­bi­li­ty and flexibility.

AMD is exhi­bi­ting with AMD Fire­Pro S9150 ser­ver GPUs and AMD ser­ver tech­no­lo­gies at SC ’14 in New Orleans Ernest Memo­ri­al Con­ven­ti­on Cen­ter booth #839 from Novem­ber 17 to 20.

Supporting Resources

About AMD

AMD (NYSE: AMD) is a semi­con­duc­tor design inno­va­tor lea­ding the next era of vivid digi­tal expe­ri­en­ces with its ground­brea­king AMD Acce­le­ra­ted Pro­ces­sing Units (APUs) that power a wide ran­ge of com­pu­ting devices. AMD’s ser­ver com­pu­ting pro­ducts are focu­sed on dri­ving indus­try-lea­ding cloud com­pu­ting and vir­tua­liza­ti­on envi­ron­ments. AMD’s supe­ri­or gra­phics tech­no­lo­gies are found in a varie­ty of solu­ti­ons ran­ging from game con­so­les, PCs to super­com­pu­ters. For more infor­ma­ti­on, visit www.amd.com.

—30—

AMD, the AMD Arrow logo and Fire­Pro are trade­marks of Advan­ced Micro Devices, Inc. Other names are for infor­ma­tio­nal pur­po­ses only and may be trade­marks of their respec­ti­ve owners.OpenCL and the Open­CL logo are trade­marks of Apple Inc. used by per­mis­si­on by Khronos.

 

1 The first 60 nodes at GSI are alre­a­dy acti­ve (240 S9150 = 1.22 peta­FLOPS peak GPU sin­gle pre­cis­i­on and 610 tera­FLOPS GPU dou­ble pre­cis­i­on). By the end of the year, the remai­ning 100 nodes will be imple­men­ted (400 S9150 = 2.03 peta­FLOPS peak GPU sin­gle pre­cis­i­on and 1.01 peta­FLOPS peak GPU dou­ble precision)