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.

Penguin Computing’s Cluster Collaboration with Advanced Micro Devices Makes Heterogeneous System Architecture Clustering a Reality

Verfasst vonDr@

Hete­ro­ge­neous Sys­tem Archi­tec­tu­re Pro­gress Pro­vi­des Indus­try Lea­der­ship for More Effi­ci­ent Computing

New Orleans, LA-SC14-Nov. 20, 2014 — Pen­gu­in Com­pu­ting, pro­vi­der of high per­for­mance com­pu­ting, enter­pri­se data cen­ter and cloud solu­ti­ons, today announ­ced the first appli­ca­ti­on opti­mi­zed acce­le­ra­ted pro­ces­sing unit (APU) clus­ters, making seam­less GPU and CPU memo­ry sha­ring on clus­ters a rea­li­ty based on hete­ro­ge­neous sys­tem archi­tec­tu­re (HSA) from Advan­ced Micro Devices (NYSE: AMD). The shared memo­ry capa­bi­li­ty invol­ves very light­weight con­text swit­ches to switch instanta­neous­ly bet­ween the GPU and CPU, whi­che­ver code runs best at a given moment.

Today’s appli­ca­ti­ons requi­re power effi­ci­en­cy and high per­for­mance achie­ved through high­ly par­al­lel com­pu­ting, howe­ver GPUs and CPUs pre­sent a bot­t­len­eck to effi­ci­ent clus­ter func­tion­a­li­ty due to sepa­ra­te memo­ry space. This tra­di­tio­nal archi­tec­tu­re results in inef­fi­ci­ent GPU/CPU com­mu­ni­ca­ti­on, which is a chall­enge to scaling.

“We are making the­se machi­nes imme­dia­te­ly available for eva­lua­ti­on as a tre­men­dous tool for soft­ware deve­lo­p­ment,” said Phil Pokor­ny, Chief Tech­no­lo­gy Offi­cer, Pen­gu­in Com­pu­ting. “HSA is a rea­li­ty and our tech­no­lo­gy is alre­a­dy in the hands of major U.S. labs. Pen­gu­in Computing’s exten­si­ve expe­ri­ence in APU clus­ter deve­lo­p­ment and imple­men­ta­ti­on is instru­men­tal in this pro­gress, in addi­ti­on to clo­se col­la­bo­ra­te with AMD.”

Named Jää­tik­kö, or ice­berg in Fin­nish, the clus­ter is curr­ent­ly being demons­tra­ted at AMD’s SC14 booth #839 and com­bi­nes 10 AMD APU com­pu­te nodes, plus head node based on Penguin’s Alt­us 2A30 deve­lo­p­ment plat­form, with high per­for­mance ether­net using Penguin’s Arc­ti­ca open ether­net switches.

“Initi­al feed­back from ear­ly adop­ters rein­forces our belief that this col­la­bo­ra­ti­on with Pen­gu­in Com­pu­ting is an important step for­ward for the indus­try,” said Karl Freund, cor­po­ra­te vice pre­si­dent, Pro­duct Manage­ment and Mar­ket, Ser­ver Busi­ness Unit, AMD. “The poten­ti­al of modern hete­ro­ge­neous archi­tec­tures is exci­ting, and col­la­bo­ra­ti­ons such as the­se can result in signi­fi­cant steps for­ward in per­for­mance for a broad ran­ge of soft­ware applications.”

The oil and gas indus­try is an exam­p­le of a cus­to­mer seg­ment that could expe­ri­ence signi­fi­cant bene­fits from this capa­bi­li­ty. With the oil and gas sector’s need for GPU par­al­lel codes and sin­gle-pre­cis­i­on, APU has almost a tera­flop of sin­gle pre­cis­i­on floa­ting point performance.

The com­bi­ned Pen­gu­in Com­pu­ting and AMD solu­ti­on is a very cost-effec­ti­ve way to get star­ted with one node or cluster.

Visit http://www.penguincomputing.com/products/rackmount-servers/altus/altus-2a30/ to learn more about the Alt­us 2a30.