SUPERCOMPUTING LEADER CRAY INTRODUCES FIRST EXASCALE-CLASS SUPERCOMPUTER

Cray Sets New Stan­dard for Indus­try; Natio­nal Ener­gy Rese­arch Sci­en­ti­fic Com­pu­ting (NERSC) Cen­ter Taps Cray Sys­tem for 2020

SEATTLE, Oct. 30, 2018 (GLOBE NEWSWIRE) — Glo­bal super­com­pu­ter lea­der Cray Inc. (Nasdaq:CRAY) today unvei­led its new, revo­lu­tio­na­ry super­com­pu­ting sys­tem code-named “Shas­ta™.” The sys­tem will be show­ca­sed next month at the 30th anni­ver­s­a­ry of the Super­com­pu­ting (SC) con­fe­rence in Dal­las, Texas, known as SC18. Shas­ta is an ent­i­re­ly new design and is set to be the tech­no­lo­gy that under­pins the next era of super­com­pu­ting, cha­rac­te­ri­zed by exas­ca­le per­for­mance capa­bi­li­ty, new data-centric workloads, and an explo­si­on of pro­ces­sor archi­tec­tures.  With swee­ping hard­ware and soft­ware inno­va­tions, Shas­ta incor­po­ra­tes next-genera­ti­on Cray sys­tem soft­ware to enab­le modu­la­ri­ty and exten­si­bi­li­ty, a new Cray-desi­gned sys­tem inter­con­nect, unpar­al­leled fle­xi­bi­li­ty in pro­ces­sing choice wit­hin a sys­tem, and a soft­ware envi­ron­ment that pro­vi­des for seam­less sca­la­bi­li­ty. The U.S. Depart­ment of Ener­gy (DOE) also announ­ced today that NERSC, the Natio­nal Ener­gy Rese­arch Sci­en­ti­fic Com­pu­ting Cen­ter, has cho­sen a Cray “Shas­ta” super­com­pu­ter for its NERSC-9 sys­tem, named “Perl­mut­ter,” in 2020.  The pro­gram con­tract is valued at $146 mil­li­on, one of the lar­gest in Cray’s histo­ry, and will fea­ture a Shas­ta sys­tem with Cray® Clus­terStor™ sto­rage.

Super­com­pu­ting Rede­si­gned

Cray super­com­pu­ter sys­tems con­sistent­ly lead the indus­try in per­for­mance and effi­ci­ent sca­ling. Shas­ta con­ti­nues this lea­dership into much lar­ger capa­bi­li­ty sys­tems, star­ting from a sin­gle cabi­net and sca­ling up to exas­ca­le and bey­ond, to ans­wer more ques­ti­ons in all fields in sup­port of extre­me-sca­le sci­ence, inno­va­ti­on and dis­co­very.  At a time when demand is rising for sin­gle sys­tems to hand­le con­ver­ged mode­ling, simu­la­ti­on, AI and ana­ly­tics workloads, Shasta’s data-centric design allows it to run diver­se workloads and work­flows all on one sys­tem, all at the same time.  Shasta’s hard­ware and soft­ware inno­va­tions tack­le the bot­t­len­ecks, mana­gea­bi­li­ty, and job com­ple­ti­on issu­es that emer­ge or are magni­fied as core counts grow, com­pu­te node archi­tec­tures pro­li­fe­ra­te, and work­flows expand to incor­po­ra­te AI at sca­le.  Shas­ta eli­mi­na­tes the dis­tinc­tion bet­ween clus­ters and super­com­pu­ters with a sin­gle new bre­akth­rough super­com­pu­ting sys­tem archi­tec­tu­re, enab­ling custo­mers to choo­se the com­pu­ta­tio­nal infra­st­ruc­tu­re that best fits their mis­si­on, wit­hout tra­de­offs.  With Shas­ta you can mix and match pro­ces­sor archi­tec­tures (X86, Arm®, GPUs) in the same sys­tem as well as sys­tem inter­con­nec­ts from Cray (Slingshot™), Intel (Omni-Path) or Mel­lanox (Infi­ni­Band®).

Shas­ta will usher in a new era of super­com­pu­ting and rep­res­ents a true game-chan­ger at a time when arti­fi­ci­al intel­li­gence and ana­ly­tics are being brought to bear on increa­singly lar­ge and com­plex pro­blems, inclu­ding clas­sic HPC mode­ling and simu­la­ti­on chal­len­ges, across an ever-broa­de­ning set of com­pa­nies and indus­tries,” said Peter Unga­ro, pre­si­dent and CEO of Cray.  “It is also very exci­ting to announ­ce that one of the lar­gest con­trac­ts in the histo­ry of our com­pa­ny was just signed with NERSC.  We are hono­red to con­ti­nue our part­nership with NERSC and put Shas­ta to work in sup­port of their broad mis­si­on to enab­le com­pu­ta­tio­nal and data sci­ence at sca­le.”

Our sci­en­tists gather mas­si­ve amounts of data from sci­en­ti­fic instru­ments like tele­scopes and detec­tors that our super­com­pu­ters ana­ly­ze every day,” said Dr. Sudip Dosan­jh, direc­tor of the NERSC Cen­ter at Law­rence Ber­ke­ley Natio­nal Labo­ra­to­ry.  “The Shas­ta system’s ease of use and adap­ta­bi­li­ty to modern work­flows and app­li­ca­ti­ons will allow us to broa­den access to super­com­pu­ting and enab­le a who­le new pool of users.  The abi­li­ty to bring this data into the super­com­pu­ter will allow us to quick­ly and effi­ci­ent­ly sca­le and redu­ce over­all time to dis­co­very.  We value being able to work clo­se­ly with Cray to pro­vi­de our feed­back on this next genera­ti­on sys­tem which is so cri­ti­cal to exten­ding our Center’s inno­va­ti­on.”

Cray is wide­ly seen as one of only a few HPC ven­dors world­wi­de that is capa­ble of aggres­si­ve tech­no­lo­gy inno­va­ti­on at the sys­tem archi­tec­tu­re level,” said Ste­ve Con­way, Hype­ri­on Rese­arch seni­or vice pre­si­dent of rese­arch.  “Cray’s Shas­ta archi­tec­tu­re clo­se­ly matches the wish list that lea­ding HPC users have for the exas­ca­le era, but didn’t expect to be avail­ab­le this soon.  This is tru­ly a bre­akth­rough achie­ve­ment.”

Cray Slingshot Net­work – Desi­gned for Data-Centric Com­pu­ting

With Shas­ta, Cray is also announ­cing Slingshot, a new high-speed, pur­po­se-built super­com­pu­ting inter­con­nect.  Slingshot advan­ces Cray’s indus­try lea­dership in scala­b­le net­work per­for­mance and adds capa­bi­li­ties that broa­den Cray’s mar­ket reach.  The Cray-deve­lo­ped Slingshot inter­con­nect will have up to 5x more band­width per node and is desi­gned for data-centric com­pu­ting.  Slingshot will fea­ture Ether­net com­pa­ti­bi­li­ty, advan­ced adap­ti­ve rou­ting, first-of-a-kind con­ges­ti­on con­trol, and sophisti­ca­ted qua­li­ty-of-ser­vice capa­bi­li­ties.  Sup­port for both IP-rou­t­ed and remo­te memo­ry ope­ra­ti­ons will broa­den the ran­ge of app­li­ca­ti­ons bey­ond tra­di­tio­nal mode­ling and simu­la­ti­on.  Qua­li­ty-of-ser­vice and novel con­ges­ti­on manage­ment fea­tures will limit the impact to cri­ti­cal workloads from sys­tem ser­vices, I/O traf­fic, and co-ten­ant workloads, to increa­se rea­li­zed per­for­mance and limit per­for­mance varia­ti­on.  Reduc­tion in the net­work dia­me­ter from five hops (in the cur­rent Cray® XC™ genera­ti­on) to three will redu­ce laten­cy and power while impro­ving sustai­ned band­width and relia­bi­li­ty.

We lis­tened clo­se­ly to our custo­mers and dug into the future needs of AI and HPC app­li­ca­ti­ons as we desi­gned Shas­ta,” said Ste­ve Scott, seni­or vice pre­si­dent and CTO of Cray.  “Custo­mers wan­ted lea­ding-edge, scala­b­le per­for­mance, but with lots of fle­xi­bi­li­ty and easy upgradea­bi­li­ty over time.  I’m hap­py to say we’ve nai­led this with Shas­ta.  The Shas­ta infra­st­ruc­tu­re accom­mo­da­tes a wide varie­ty of pro­ces­sor and net­work opti­ons, allo­wing custo­mers to run diver­se workloads on a sin­gle sys­tem.  And it’s got the headroom to accom­mo­da­te increa­singly power-hungry pro­ces­sors and acce­le­ra­tors com­ing in future years.  The Slingshot net­work tight­ly binds the com­pu­te and sto­rage resour­ces in the sys­tem, with ground­brea­king con­ges­ti­on con­trol to iso­la­te app­li­ca­ti­ons from other net­work traf­fic, and Ether­net com­pa­ti­bi­li­ty for dat­a­cen­ter and sto­rage inte­gra­ti­on.  We’re immen­se­ly exci­ted to bring this new net­work to mar­ket to help acce­le­ra­te our custo­mers’ dis­co­ve­ries.”

Fle­xi­bi­li­ty

Shas­ta lets custo­mers ful­ly rea­li­ze Cray’s long­time visi­on of adap­ting super­com­pu­ting sys­tems to workloads using opti­mi­zed pro­ces­sing and net­wor­king. This beco­mes addi­tio­nal­ly valu­able as custo­mers are increa­singly con­cer­ned about choo­sing the opti­mal archi­tec­tu­re as their workloads rapidly evol­ve.  With Shas­ta, Cray can incor­po­ra­te any pro­ces­sor choice — or a hete­ro­ge­ne­ous mix — with a sin­gle manage­ment and app­li­ca­ti­on deve­lop­ment infra­st­ruc­tu­re.  Custo­mers can flex from sin­gle to mul­ti-socket pro­ces­sor nodes, GPUs, FPGAs and other pro­ces­sing opti­ons that will emer­ge, such as AI spe­cia­li­zed acce­le­ra­tors.  Custo­mers can make late-bin­ding decisi­ons on com­pu­te tech­no­lo­gy and not sacri­fice capa­bi­li­ty, becau­se Shasta’s design allows tailo­ring of sys­tem den­si­ty and injec­tion band­width to opti­mi­ze pri­ce and per­for­mance.

Cray’s Part­ners Express Sup­port of Shas­ta

AMD:
“We are tre­men­dous­ly proud that AMD EPYC™ pro­ces­sors are part of the next-genera­ti­on Cray Shas­ta plat­form at NERSC,” said For­rest Nor­rod, seni­or vice pre­si­dent and gene­ral mana­ger, AMD Dat­a­cen­ter and Embed­ded Sys­tems Group.  “The Shas­ta sys­tem rep­res­ents a new para­digm for super­com­pu­ting that is going to take the indus­try into the exas­ca­le era.  The Cray Shas­ta archi­tec­tu­re sup­ports both AMD EPYC CPUs and acce­le­ra­tors, inclu­ding AMD Rade­on™ Instinct™ GPUs.  That com­bi­na­ti­on of CPU and GPU power offers an incredi­ble level of per­for­mance for many HPC app­li­ca­ti­ons.” 

Intel:
“Intel and Cray have a long-stan­ding stra­te­gic part­nership to deli­ver super­com­pu­ting sys­tems. We are exci­ted to work with Cray to advan­ce super­com­pu­ting capa­bi­li­ty with their new Shas­ta plat­form opti­mi­zed for Intel® Xeon® Scala­b­le pro­ces­sors and other Intel HPC pro­duc­ts.  Intel’s sui­te of HPC pro­duc­ts pro­vi­des the buil­ding blocks for a balan­ced design that deli­vers the per­for­mance and sca­la­bi­li­ty nee­ded to tack­le the extre­me chal­len­ges of both HPC and AI on a com­mon infra­st­ruc­tu­re,” said Trish Dam­kro­ger, vice pre­si­dent and gene­ral mana­ger, Intel’s Data Cen­ter Group.

Mar­vell:
“Mar­vell and Cray have a long-stan­ding rela­ti­ons­hip of col­la­bo­ra­ti­on to bring Arm-based pro­ces­sors to the high-per­for­mance com­pu­ting envi­ron­ment,” said Rag­hib Hus­sain, exe­cu­ti­ve vice pre­si­dent and chief stra­te­gy offi­cer, Mar­vell Semi­con­duc­tor, Inc.  “The Cray XC50 super­com­pu­ter, now com­mer­ci­al­ly avail­ab­le, inte­gra­tes Mar­vell® ThunderX2®processors based on 64-bit Arm­v8-A archi­tec­tu­re to bring best-in-class SoC com­pu­te per­for­mance and memo­ry band­width.  We congra­tu­la­te Cray on the intro­duc­tion of its exas­ca­le-class Shas­ta sys­tem and look for­ward to con­ti­nuing to work with Cray to usher in the next era of Arm-based super­com­pu­ting.” 

Mel­lanox:
“We are hap­py to sup­port the new Shas­ta plat­form with our smart Infi­ni­Band inter­con­nect solu­ti­ons, enab­ling our mutu­al custo­mers to achie­ve lea­ding per­for­mance and sca­la­bi­li­ty,” said Gilad Shai­ner, vice pre­si­dent of mar­ke­ting at Mel­lanox Tech­no­lo­gies.  “The com­bi­na­ti­on of InfiniBand’s acce­le­ra­ti­on engi­nes and Cray’s soft­ware will dri­ve the next genera­ti­on of HPC and AI app­li­ca­ti­ons and workloads.”

NVIDIA:
“The Shas­ta-based sys­tem selec­ted for NERSC’s new Perl­mut­ter super­com­pu­ter incor­po­ra­tes the most-advan­ced NVIDIA GPU acce­le­ra­ti­on plat­form avail­ab­le, pro­vi­ding the U.S. Depart­ment of Ener­gy a clear path to exas­ca­le com­pu­ting,” said Ian Buck, vice pre­si­dent and gene­ral mana­ger of acce­le­ra­ted com­pu­ting at NVIDIA. “Whe­ther used for tra­di­tio­nal sci­en­ti­fic com­pu­ting or new, emer­ging AI and machi­ne lear­ning workloads, Perl­mut­ter will offer the sci­en­tists that use it an incredi­b­ly advan­ced acce­le­ra­ti­on plat­form, an essen­ti­al tool to help them unlock mys­te­ries and advan­ce sci­ence.”
            
Super­char­ged TCO

Desi­gned from the ground up to sup­port a deca­de or more of advan­ce­ments in com­pu­ta­tio­nal pro­ces­sing, Shas­ta eli­mi­na­tes the need for fre­quent, expen­si­ve upgrades and will dri­ve excep­tio­nal­ly low total cost of ownership over the life­time of the sys­tem.  Shas­ta packa­ging comes in two opti­ons: a 19” air- or liquid-coo­led, stan­dard dat­a­cen­ter rack and a high-den­si­ty, liquid-coo­led rack desi­gned to hold 64 com­pu­te bla­des with mul­ti­ple pro­ces­sors per bla­de.  Both opti­ons can sca­le to well over 100 cabi­nets.  As pro­ces­sor wat­ta­ge increa­ses over time to boost com­pu­ta­tio­nal per­for­mance, Shasta’s fle­xi­ble coo­ling design eli­mi­na­tes the need to do for­klift upgrades of sys­tem infra­st­ruc­tu­re to accom­mo­da­te hig­her-power pro­ces­sors.  Cray desi­gned Shas­ta to sup­port pro­ces­sors excee­ding 500 watts with high­ly effi­ci­ent coo­ling, resul­ting in less was­te and lower costs while mee­ting cri­ti­cal pro­ces­sor den­si­ty requi­re­ments.  Shas­ta sys­tems are also desi­gned to meet warm-water-coo­ling dat­a­cen­ter stan­dards like W3 and W4 requi­re­ments throughout the world.  

Avai­la­bi­li­ty

Shas­ta sys­tems are expec­ted to be com­mer­ci­al­ly avail­ab­le in Q4 of 2019.  Plea­se con­tact your aut­ho­ri­zed Cray sales rep­re­sen­ta­ti­ve for more infor­ma­ti­on.

SC18

Stop by the Cray booth at SC18: The Inter­na­tio­nal Con­fe­rence for High Per­for­mance Com­pu­ting, Net­wor­king, Sto­rage, and Ana­ly­sis, Novem­ber 11–16, 2018 in Dal­las, Texas to learn more about Shas­ta and Cray’s new inter­con­nect, Slingshot.

About Cray Inc.
Cray Inc. (Nasdaq:CRAY) com­bi­nes com­pu­ta­ti­on and crea­ti­vi­ty so visio­na­ries can keep asking ques­ti­ons that chal­len­ge the limits of pos­si­bi­li­ty.  Drawing on more than 45 years of expe­ri­ence, Cray deve­lops the world’s most advan­ced super­com­pu­ters, pushing the bounda­ries of per­for­mance, effi­ci­en­cy and sca­la­bi­li­ty.  Cray con­ti­nues to inno­va­te today at the con­ver­gence of data and dis­co­very, offe­ring a com­pre­hen­si­ve port­fo­lio of super­com­pu­ters, high-per­for­mance sto­rage, data ana­ly­tics and arti­fi­ci­al intel­li­gence solu­ti­ons.  Go to www.cray.com for more infor­ma­ti­on.

Safe Har­bor State­ment
This press release con­tains for­ward-loo­king state­ments wit­hin the mea­ning of Sec­tion 21E of the Secu­ri­ties Exchan­ge Act of 1934 and Sec­tion 27A of the Secu­ri­ties Act of 1933, inclu­ding, but not limi­ted to, state­ments rela­ted to Cray’s pro­duct deve­lop­ment, sales, and deli­very plans rela­ting to Shas­ta super­com­pu­ting sys­tems, the timing and deli­very of the com­pu­te and sto­rage sys­tem orde­red by NERSC and Cray’s abi­li­ty to deli­ver a sys­tem that meets NERSC’s requi­re­ments.  The­se state­ments invol­ve cur­rent expec­ta­ti­ons, fore­casts of future events and other state­ments that are not his­to­ri­cal fac­ts. Inac­cu­ra­te assump­ti­ons and known and unknown risks and uncer­tain­ties can affect the accu­ra­cy of for­ward-loo­king state­ments and cau­se actu­al results to dif­fer mate­ri­al­ly from tho­se anti­ci­pa­ted by the­se for­ward-loo­king state­ments.  Fac­tors that could affect actu­al future events or results inclu­de, but are not limi­ted to, the risk that Cray is not able to suc­cess­ful­ly com­ple­te its plan­ned pro­duct deve­lop­ment efforts or to ship Shas­ta sys­tems wit­hin the plan­ned time­frame or at all, the risk that Shas­ta sys­tems will not have the fea­tures, per­for­mance or com­pon­ents cur­r­ent­ly plan­ned, the risk that pro­ces­sors and inter­con­nect tech­no­lo­gy plan­ned for Cray Shas­ta sys­tems are not avail­ab­le when expec­ted or with the per­for­mance or pri­cing expec­ted, the risk that the com­pu­te and sto­rage sys­tem requi­red by NERSC is not deli­ve­r­ed in a time­ly fashion or does not per­form as expec­ted and such other risks as iden­ti­fied in the Company’s quar­ter­ly report on Form 10-Q for the quar­ter ended June 30, 2018, and from time to time in other reports filed by Cray with the U.S. Secu­ri­ties and Exchan­ge Com­mis­si­on.  You should not rely undu­ly on the­se for­ward-loo­king state­ments, which app­ly only as of the date of this release.  Cray under­ta­kes no duty to publicly announ­ce or report revi­si­ons to the­se state­ments as new infor­ma­ti­on beco­mes avail­ab­le that may chan­ge the Company’s expec­ta­ti­ons.

CRAY is a regis­te­red trade­mark of Cray Inc. in the United Sta­tes and other coun­tries, and Clus­terStor, Slingshot and XC are trade­marks of Cray Inc.  Other pro­duct and ser­vice names men­tio­ned her­ein are the trade­marks of their respec­tive owners.