SANTA CLARA, Calif.—Oct. 29, 2015—NVIDIA today announ­ced avai­la­bi­li­ty of ver­si­on 15.10 of the PGI Acce­le­ra­tor™ For­tran, C and C++ com­pi­lers, adding sup­port for the Open­ACC® direc­ti­ves-based par­al­lel pro­gramming stan­dard on x86 archi­tec­tu­re mul­ti­co­re microprocessors.

The new PGI com­pi­lers deli­ver per­for­mance por­ta­bi­li­ty, allo­wing Open­ACC-enab­led source code to be com­pi­led for par­al­lel exe­cu­ti­on on a mul­ti­co­re CPU or a GPU acce­le­ra­tor. This capa­bi­li­ty pro­vi­des tre­men­dous fle­xi­bi­li­ty for pro­gramm­ers, enab­ling them to deve­lop appli­ca­ti­ons that can take advan­ta­ge of mul­ti­ple sys­tem archi­tec­tures with a sin­gle ver­si­on of their source code.

“Our goal is to enable HPC deve­lo­pers to easi­ly port appli­ca­ti­ons across all major CPU and acce­le­ra­tor plat­forms with uni­form­ly high per­for­mance using a com­mon source code base,” said Dou­glas Miles, direc­tor of PGI Com­pi­lers & Tools at NVIDIA. “This capa­bi­li­ty will be par­ti­cu­lar­ly important in the race towards exas­ca­le com­pu­ting in which the­re will be a varie­ty of sys­tem archi­tec­tures requi­ring a more fle­xi­ble appli­ca­ti­on pro­gramming approach.”

This new PGI fea­ture com­pi­les Open­ACC com­pu­te regi­ons for par­al­lel exe­cu­ti­on across all of the cores in an x86 pro­ces­sor or mul­ti-socket ser­ver. The cores are trea­ted in aggre­ga­te as a shared-memo­ry acce­le­ra­tor, eli­mi­na­ting all data move­ment over­head in the resul­ting Open­ACC pro­grams. By default the com­pi­ler gene­ra­tes code that uses all the available cores in the sys­tem, and seve­ral methods exist for pro­gramm­ers to con­trol and fine-tune this behavior.

Key bene­fits of run­ning Open­ACC on mul­ti­co­re CPUs include:

• Effec­ti­ve uti­liza­ti­on of all cores of a mul­ti­co­re CPU or mul­ti-socket ser­ver for par­al­lel execution
• Com­mon pro­gramming model across CPUs and GPUs in For­tran, C and C++
• Rapid explo­ita­ti­on of exis­ting mul­ti­co­re par­al­le­lism in a pro­gram using the KERNELS direc­ti­ve, which enables incre­men­tal opti­miza­ti­on for par­al­lel execution
• Sca­lable per­for­mance across mul­ti­co­re CPUs and GPUs

“Port­ing HPC appli­ca­ti­ons from one plat­form to ano­ther is one of the most signi­fi­cant cos­ts in the adop­ti­on of breakth­rough hard­ware tech­no­lo­gies,” said Bud­dy Bland, pro­ject direc­tor at Oak Ridge Natio­nal Labo­ra­to­ry. “Open­ACC for mul­ti­co­re x86 CPUs pro­vi­des con­ti­nui­ty and code por­ta­bi­li­ty from exis­ting CPU-only and GPU-enab­led appli­ca­ti­ons from machi­nes like Titan to all of DOE’s upco­ming major sys­tems as well as por­ta­bi­li­ty among tho­se systems.”

Gro­wing Momen­tum for OpenACC

The­re are more than 10,000 deve­lo­pers using Open­ACC today, and seve­ral recent deve­lo­p­ments unders­core the con­ti­nu­al­ly gro­wing adop­ti­on of Open­ACC in high per­for­mance com­pu­ting. At recent hacka­thons con­duc­ted world­wi­de, experts across a varie­ty of sci­en­ti­fic domains have been acce­le­ra­ting their sci­en­ti­fic appli­ca­ti­ons with acce­le­ra­tors and Open­ACC. The­se include appli­ca­ti­ons in such diver­se fields as MRI image recon­s­truc­tion (PowerG­rid), com­pu­ta­tio­nal flu­id dyna­mics (INCOMP3D, HiP­STAR and Nume­ca), cos­mo­lo­gy and astro­phy­sics (RAMSES, CASTRO and MAESTRO), quan­tum che­mis­try (LSDALTON), com­pu­ta­tio­nal phy­sics (Nek­CEM) and more.

In addi­ti­on, Gaus­si­an, Inc. has announ­ced that it is using Open­ACC to port the GAUSSIAN com­pu­ta­tio­nal che­mis­try appli­ca­ti­on to acce­le­ra­tors. At the recent iCAS2 con­fe­rence on cli­ma­te and wea­ther in Anne­cy, France, Meteo­su­is­se, the Swiss Fede­ral Office of Meteo­ro­lo­gy and Cli­ma­to­lo­gy, announ­ced the deploy­ment of a GPU-acce­le­ra­ted ver­si­on of COSMO, the world’s first pro­duc­tion wea­ther fore­cas­ting appli­ca­ti­on run­ning on GPU accelerators.

In a recent poll of 150 Open­ACC deve­lo­pers, 94 per­cent of the respond­ents repor­ted get­ting a spee­dup when run­ning on an acce­le­ra­tor, and over 90 per­cent of the users would recom­mend OpenACC.

More infor­ma­ti­on on the PGI Acce­le­ra­tor com­pi­lers with Open­ACC sup­port is available at www.pgroup.com/accel. More infor­ma­ti­on on the Open­ACC API and stan­dard is at www.openacc.org.

In addi­ti­on to sup­port for Open­ACC on mul­ti­co­re CPUs, PGI ver­si­on 15.10 includes a pre-pro­duc­tion release of the PGI For­tran, C and C++ com­pi­lers for Open­POWER CPUs with sup­port for Open­ACC on NVIDIA GPUs.

Avai­la­bi­li­ty and Free Trial

PGI 15.10 with sup­port for Open­ACC on mul­ti­co­re CPUs is expec­ted to be available this month direct­ly from PGI and aut­ho­ri­zed resel­lers. New users can regis­ter for a free 90-day tri­al as part of the NVIDIA Open­ACC Tool­kit. Uni­ver­si­ty stu­dents and facul­ty can app­ly for a free PGI license.

About PGI Software

An NVIDIA Cor­po­ra­ti­on brand, PGI soft­ware includes high-per­for­mance par­al­lel For­tran, C and C++ com­pi­lers and tools for work­sta­tions, ser­vers and clus­ters based on x64 pro­ces­sors from Intel and AMD, and HPC acce­le­ra­tors from NVIDIA and AMD. More infor­ma­ti­on is available at www.pgroup.com, sales@pgroup.com or by cal­ling (503) 682‑2806.

Keep Cur­rent on NVIDIA

Sub­scri­be to the NVIDIA blog, fol­low us on Face­book, Goog­le+, Twit­ter, Lin­ke­dIn and Insta­gram, and view NVIDIA vide­os on You­Tube and images on Flickr.

About NVIDIA
Sin­ce 1993, NVIDIA (NASDAQ: NVDA) has pio­nee­red the art and sci­ence of visu­al com­pu­ting. The company’s tech­no­lo­gies are trans­forming a world of dis­plays into a world of inter­ac­ti­ve dis­co­very — for ever­yo­ne from gamers to sci­en­tists, and con­su­mers to enter­pri­se cus­to­mers. More infor­ma­ti­on at http://nvidianews.nvidia.com/ and http://blogs.nvidia.com/.

# # #

Cer­tain state­ments in this press release inclu­ding, but not limi­t­ed to, state­ments as to: the fea­tures, bene­fits and avai­la­bi­li­ty of PGI Acce­le­ra­tor com­pi­lers with added sup­port for Open­ACC are for­ward-loo­king state­ments that are sub­ject to risks and uncer­tain­ties that could cau­se results to be mate­ri­al­ly dif­fe­rent than expec­ta­ti­ons. Important fac­tors that could cau­se actu­al results to dif­fer mate­ri­al­ly include: glo­bal eco­no­mic con­di­ti­ons; our reli­ance on third par­ties to manu­fac­tu­re, assem­ble, packa­ge and test our pro­ducts; the impact of tech­no­lo­gi­cal deve­lo­p­ment and com­pe­ti­ti­on; deve­lo­p­ment of new pro­ducts and tech­no­lo­gies or enhance­ments to our exis­ting pro­duct and tech­no­lo­gies; mar­ket accep­tance of our pro­ducts or our part­ners’ pro­ducts; design, manu­fac­tu­ring or soft­ware defects; chan­ges in con­su­mer pre­fe­ren­ces or demands; chan­ges in indus­try stan­dards and inter­faces; unex­pec­ted loss of per­for­mance of our pro­ducts or tech­no­lo­gies when inte­gra­ted into sys­tems; as well as other fac­tors detail­ed from time to time in the reports NVIDIA files with the Secu­ri­ties and Exch­an­ge Com­mis­si­on, or SEC, inclu­ding its Form 10‑Q for the fis­cal peri­od ended July 26, 2015. Copies of reports filed with the SEC are pos­ted on the company’s web­site and are available from NVIDIA wit­hout char­ge. The­se for­ward-loo­king state­ments are not gua­ran­tees of future per­for­mance and speak only as of the date hereof, and, except as requi­red by law, NVIDIA dis­claims any obli­ga­ti­on to update the­se for­ward-loo­king state­ments to reflect future events or circumstances.

© 2015 NVIDIA Cor­po­ra­ti­on. All rights reser­ved. NVIDIA, the NVIDIA logo, PGI Acce­le­ra­tor and Open­ACC are trade­marks and/or regis­tered trade­marks of NVIDIA Cor­po­ra­ti­on in the U.S. and other count­ries. Other com­pa­ny and pro­duct names may be trade­marks of the respec­ti­ve com­pa­nies with which they are asso­cia­ted. Fea­tures, pri­cing, avai­la­bi­li­ty and spe­ci­fi­ca­ti­ons are sub­ject to chan­ge wit­hout notice.