Sys­tem fea­tures 5184 CPU cores and 20.7 TBytes of shared memo­ry, hand­les mas­si­ve amounts of data as the first part of a new lar­ge cloud com­pu­ting faci­li­ty for Big Data Analytics.

New Orleans, Novem­ber 18, 2014

Numas­ca­le today announ­ced the suc­cessful instal­la­ti­on of a lar­ge shared memo­ry Numascale/Supermicro/AMD sys­tem at a cus­to­mer dat­a­cen­ter faci­li­ty in North Ame­ri­ca. The sys­tem is the first part of a lar­ge cloud com­pu­ting faci­li­ty for ana­ly­tics and simu­la­ti­on of sen­sor data com­bi­ned with his­to­ri­cal data.

The Numas­ca­le sys­tem, instal­led over the last two weeks, con­sists of 108 Super­mi­cro 1U ser­vers con­nec­ted in a 3D torus with Numa­Connect™, using three cabi­nets with 36 ser­vers apie­ce in a 6x6x3 topo­lo­gy. Each ser­ver has 48 cores three AMD Opte­ron™ 6386 CPUs and 192 GBytes memo­ry, pro­vi­ding a sin­gle sys­tem image and 20.7 TBytes to all 5184 cores. The sys­tem was desi­gned to meet user demand for “very lar­ge memo­ry” hard­ware solu­ti­ons run­ning a stan­dard sin­gle image Linux OS on com­mo­di­ty x86 based servers.

“Supermicro’s 1U 4‑way ser­vers (AS-1042G-LTF) with Hyper­Trans­port (HTX) con­nec­ti­vi­ty pro­vi­de users with fle­xi­ble com­pu­ting resour­ces, inclu­ding capa­bi­li­ties for hand­ling very lar­ge data sets such as tho­se found in appli­ca­ti­ons for hand­ling near real-time sen­sor inputs, com­bi­ned with mas­si­ve amounts of his­to­ri­cal data,” said Tau Leng, Vice Pre­si­dent of HPC at Super­mi­cro. “Our 4‑way sys­tems with can be used as a sin­gle sys­tem or par­ti­tio­ned into smal­ler sys­tems whe­re each par­ti­ti­on runs one ins­tance of the OS. With pro­per Numa-awa­re­ness, appli­ca­ti­ons with high band­width requi­re­ments will be able to uti­li­ze the com­bi­ned band­width of all of the memo­ry con­trol­lers and still be able to share data with low laten­cy access through the coher­ent shared memory.”

Sys­tems equip­ped with Numa­Connect pro­vi­de high per­for­mance shared memo­ry pro­gramming capa­bi­li­ties with the same cost struc­tu­re as a clus­ter. Numa­connect™ also eli­mi­na­tes the dif­fi­cul­ty of MPI coding for big data pro­blems and can increase pro­gramm­er pro­duc­ti­vi­ty. This alter­na­ti­ve repres­ents a com­pel­ling solu­ti­on for sci­en­tists who curr­ent­ly work with their shared memo­ry codes on x86 desk­tops and lap­tops. The­se users can now sca­le up their data sets wit­hout any extra effort within a fami­li­ar, stan­dard Linux OS environment.

“AMD is proud to have been sel­ec­ted to par­ti­ci­pa­te in this exci­ting pro­ject,” said Suresh Gopa­l­a­krish­n­an, Gene­ral Mana­ger and Cor­po­ra­te Vice Pre­si­dent, Ser­ver Busi­ness Unit, AMD.  “This col­la­bo­ra­ti­on with Numas­ca­le and Super­mi­cro brings tog­e­ther the right tech­no­lo­gies to sol­ve the chal­len­ging pro­blem of lar­ge shared memo­ry implementation.”

Einar Rustad, CTO of Numas­ca­le, agrees. “The sin­gle memo­ry image clus­ter pro­vi­des both shared memo­ry — inclu­ding threads and OpenMP — and MPI pro­gramming opti­ons,” he said. “The sca­lable sys­tem takes advan­ta­ge of low cost com­mo­di­ty x86 hard­ware to offer signi­fi­cant savings when com­pared to con­ven­tio­nal shared memo­ry sys­tems. With Numa­Connect™, sys­tem admi­nis­tra­ti­on is iden­ti­cal to that of a sin­gle ser­ver becau­se the­re are no sepa­ra­te node images to main­tain and dis­tri­bu­te. We are very proud to have been part of the pro­ject that enab­led us to build this ground­brea­king sys­tem in clo­se coope­ra­ti­on with Super­mi­cro and AMD.”

Numa­Connect works with AMD Opteron™-based ser­vers and pro­vi­des up to 256 TBytes of sys­tem-wide shared memo­ry using cache cohe­ren­cy logic with a direc­to­ry-based pro­to­col that sca­les to 4096 nodes. The cache cohe­ren­cy logic is imple­men­ted in an ASIC tog­e­ther with inter­con­nect fabric cir­cuit­ry with rou­ting tables for mul­ti-dimen­sio­nal Torus topo­lo­gies. This type of fabric is very sca­lable and the same topo­lo­gy is used in the world’s lar­gest supercomputers.

Visit Numas­ca­le in booth 1923 at SC14 to see live pro­duct demos.

About Advan­ced Micro Devices, Inc. (AMD)

AMD (NYSE: AMD) designs and inte­gra­tes tech­no­lo­gy that powers mil­li­ons of intel­li­gent devices, inclu­ding per­so­nal com­pu­ters, tablets, game con­so­les, and cloud ser­vers that defi­ne the new era of sur­round com­pu­ting. AMD solu­ti­ons enable peo­p­le ever­y­whe­re to rea­li­ze the full poten­ti­al of their favo­ri­te devices and appli­ca­ti­ons to push the boun­da­ries of what is pos­si­ble. For more infor­ma­ti­on, visit www.amd.com.

About Super­mi­cro Com­pu­ter, Inc. 

Super­mi­cro® (NASDAQ: SMCI), the lea­ding inno­va­tor in high-per­for­mance, high-effi­ci­en­cy ser­ver tech­no­lo­gy, is a pre­mier pro­vi­der of advan­ced ser­ver Buil­ding Block Solu­ti­ons® for Data Cen­ter, Cloud Com­pu­ting, Enter­pri­se IT, Hadoop/Big Data, HPC, and Embedded Sys­tems world­wi­de. Super­mi­cro is com­mit­ted to pro­tec­ting the envi­ron­ment through its “We Keep IT Green®” initia­ti­ve and pro­vi­des cus­to­mers with the most ener­gy-effi­ci­ent, envi­ron­men­tal­ly-fri­end­ly solu­ti­ons available on the mar­ket. For more infor­ma­ti­on, visit www.supermicro.com.

About Numas­ca­le

With offices in Euro­pe, Asia, and the USA, Numascale’s ground­brea­king Numa­Connect inter­con­nect tech­no­lo­gy is the back­bone for the most sca­lable ser­ver com­pu­ter sys­tems. Numa­Connect allows high-volu­me manu­fac­tu­red ser­ver boards to be used as buil­ding blocks for sys­tems with fea­tures that are only found in high-end enter­pri­se ser­vers. Numa­Connect includes full sup­port for vir­tua­liza­ti­on of pro­ces­sing, memo­ry, and I/O resour­ces, and can be used with stan­dard ope­ra­ting systems.

Numas­ca­le is sup­port­ed by: Sta­toil, Pro­Ven­ture, Inves­tinor, Inno­va­ti­on Nor­way, Nor­ges forsk­nings­råd, and Euro­stars. For more infor­ma­ti­on, visit www.numascale.com.