Company’s soft­ware-defi­ned com­posable ser­ver sys­tems now lead SPEC CPU 2017 bench­mark results by gap of up to 96% vs. com­pe­ting solutions

FORT LEE, N.J.–(BUSINESS WIRE)–Scal­eMP™, a lea­ding pro­vi­der of vir­tua­liza­ti­on solu­ti­ons for high-end com­pu­ting, today announ­ced that it has accom­plished best-ever recor­ded results for the SPEC CPU® 2017 bench­mark for any sys­tem size of four pro­ces­sors and hig­her. ScaleMP’s results for 32‑, 24- and 16-pro­ces­sor sys­tems are now hol­ding the top three posi­ti­ons for best results ever.

In 2019, AMD and Scal­eMP announ­ced a stra­te­gic col­la­bo­ra­ti­on to enable AMD ser­ver OEMs to crea­te sys­tems with four, eight and up to 128 pro­ces­sor sockets, up to 8,192 CPUs and 256 tera­bytes of shared memo­ry using ScaleMP’s vSMP Foun­da­ti­on line of pro­ducts.

“We are extre­me­ly plea­sed with the­se new bench­mark results, which show supe­rio­ri­ty across the board,” said Shai Fult­heim, foun­der and CEO of Scal­eMP. “Cus­to­mers can now sca­le their sys­tems on demand for more com­pu­ting or more memo­ry, pay only for what they need, and be sure that they are buy­ing the best and most cost-effec­ti­ve solu­ti­on available that deli­vers the best performance.”

“ScaleMP’s uni­que soft­ware-defi­ned ser­ver tech­no­lo­gy enables AMD to cap­tu­re the lea­ding posi­ti­ons for the sys­tem-sca­le SPEC CPU2017 bench­mark wit­hout any cus­tom engi­nee­ring,” said Rag­hu Nam­bi­ar, cor­po­ra­te vice pre­si­dent, Dat­a­cen­ter Eco­sys­tems & Appli­ca­ti­on Engi­nee­ring at AMD. “Through our col­la­bo­ra­ti­on with Scal­eMP, we expect to con­ti­nue to domi­na­te the sca­le-up bench­marks as this inno­va­ti­on continues.”

The fol­lo­wing table sum­ma­ri­zes the cur­rent best SPEC CPU2017 Inte­ger Rates and SPEC CPU2017 Floa­ting Point Rates for every sca­le of ser­ver sys­tem and con­trasts it with the pre­vious best system/result for that scale.

Scal­eMP soli­di­fies the lea­ding posi­ti­on of its paten­ted vSMP Foun­da­ti­on tech­no­lo­gy and line of pro­ducts as the best sca­le-up com­pu­ting archi­tec­tu­re for com­pu­te-inten­si­ve workloads, with up to 96% bet­ter per­for­mance com­pared to alter­na­ti­ve sys­tems. The results, made public by the SPEC.org, show that ScaleMP’s on-demand com­po­sed com­pu­ting archi­tec­tu­re pro­vi­des bet­ter sca­ling than hard­ware-only approaches.

Refe­rence Results

• SPEC CPU2017 Inte­ger Rates sor­ted by result:
  https://www.spec.org/cgi-bin/osgresults?conf=rint2017&op=fetch&sort1=BASE
• SPEC CPU2017 Floa­ting Point Rates sor­ted by result:
  https://www.spec.org/cgi-bin/osgresults?conf=rfp2017&op=fetch&sort1=BASE
• SPEC CPU2017 Inte­ger Rates sor­ted num­ber of chips result:
  https://www.spec.org/cgi-bin/osgresults?conf=rint2017&op=fetch&sort2=BASE&sort1=CHIPS&sdir1=-1
• SPEC CPU2017 Floa­ting Point Rates sor­ted num­ber of chips result:
  https://www.spec.org/cgi-bin/osgresults?conf=rfp2017&op=fetch&sort2=BASE&sort1=CHIPS&sdir1=-1

About vSMP Foundation

vSMP Foun­da­ti­on is the under­ly­ing tech­no­lo­gy for ScaleMP’s line of soft­ware-defi­ned sca­le-up pro­ducts. vSMP Ser­ve­rO­NE aggre­ga­tes mul­ti­ple, indus­try-stan­dard, off-the-shelf x86 ser­vers into a sin­gle vir­tu­al high-end sys­tem, and pro­vi­des cus­to­mers with an alter­na­ti­ve to tra­di­tio­nal, expen­si­ve, sym­me­tri­cal mul­tipro­ces­sing (SMP) sys­tems. vSMP Clus­te­rO­NE offers sim­pli­fied clus­te­ring infra­struc­tu­re with a sin­gle ope­ra­ting sys­tem. vSMP Memo­ryO­NE enables the trans­pa­rent use of NVM sto­rage as sys­tem memo­ry, to redu­ce TCO and expand bey­ond tra­di­tio­nal DRAM sca­ling limitations.

vSMP Ser­ve­rO­NE sup­ports aggre­ga­ti­on of up to 128 ser­vers into a sin­gle vir­tu­al SMP system:

• Expan­ded sys­tem memo­ry — to as much as 2PB of shared memo­ry for lar­ge-memo­ry requi­re­ments — by using DRAM from other servers
• As many as 32,768 CPUs on a vir­tu­al SMP with pro­ven sca­la­bi­li­ty and record-brea­king memo­ry bandwidth
• On-demand, com­posable SMP, with sup­port for many com­mon pro­vi­sio­ning sys­tems to enable increased data cen­ter uti­liza­ti­on with a con­so­li­da­ted architecture

vSMP Clus­te­rO­NE provides:

• The only Hyper­con­ver­ged HPTC Clus­ter infrastructure
• Ease of use and lowest total cost of owner­ship for clus­ters of up to 64 pro­ces­sor sockets
• No need for fabric or clus­ter manage­ment, cou­pled with sim­pli­fied par­al­lel sto­rage architecture

vSMP Memo­ryO­NE trans­par­ent­ly trans­forms high-end NVM devices into sys­tem memory:

• DRAM-like per­for­mance, with up to 8x memo­ry expansion
• Cost reduc­tion of up to 2.5x com­pared with DRAM-only solu­ti­ons by using enter­pri­se-gra­de non-vola­ti­le memory

For more infor­ma­ti­on on vSMP Foun­da­ti­on, plea­se visit http://www.scalemp.com/products.

About Scal­eMP

Scal­eMP is the lea­der in vir­tua­liza­ti­on for high-end com­pu­ting, pro­vi­ding increased per­for­mance and total cost of owner­ship (TCO) reduc­tion. The inno­va­ti­ve Ver­sa­ti­le SMP™ (vSMP) archi­tec­tu­re pro­vi­des soft­ware-defi­ned com­pu­ting and soft­ware-defi­ned memo­ry by aggre­ga­ting mul­ti­ple inde­pen­dent sys­tems or high-per­for­mance SSDs into sin­gle vir­tu­al sys­tems. Using soft­ware to replace cus­tom hard­ware sys­tems and com­pon­ents, Scal­eMP offers a revo­lu­tio­na­ry sca­le-up com­pu­ting para­digm by deli­ve­ring indus­try-stan­dard, high-end sym­me­tric mul­tipro­ces­sing com­pu­te and memo­ry envi­ron­ments. For more infor­ma­ti­on, plea­se call +1 (201) 429‑9740 or visit http://www.scalemp.com/.

vSMP Foun­da­ti­on is a trade­mark or regis­tered trade­mark of Scal­eMP. All other brands or pro­ducts are trade­marks or regis­tered trade­marks of their respec­ti­ve hol­ders and should be trea­ted as such.