{"id":21083,"date":"2015-11-17T20:24:33","date_gmt":"2015-11-17T19:24:33","guid":{"rendered":"http:\/\/www.planet3dnow.de\/cms\/?p=21083"},"modified":"2015-11-17T20:24:33","modified_gmt":"2015-11-17T19:24:33","slug":"amd-launches-boltzmann-initiative-to-dramatically-reduce-barriers-to-gpu-computing-on-amd-firepro-graphics","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/21083-amd-launches-boltzmann-initiative-to-dramatically-reduce-barriers-to-gpu-computing-on-amd-firepro-graphics\/","title":{"rendered":"AMD<\/span> Launches \u2018Boltzmann Initiative\u2019 to Dramatically Reduce Barriers to GPU<\/span> Computing on AMD<\/span> FirePro Graphics"},"content":{"rendered":"

Tools Tar\u00adget an Unpre\u00adce\u00adden\u00adted 28 Tera\u00adflops of Pro\u00adces\u00adsing at Less Than a Kilo\u00adwatt by 2016<\/em><\/p>\n

AUSTIN<\/span>, Texas \u2014 Novem\u00adber 16, 2015 \u2014 <\/strong>Buil\u00adding on its stra\u00adte\u00adgic invest\u00adments in hete\u00adro\u00adge\u00adneous sys\u00adtem archi\u00adtec\u00adtu\u00adre (HSA<\/span>), AMD<\/span><\/a> (NASDAQ<\/span>:AMD<\/span>) announ\u00adced a suite of tools desi\u00adgned to ease deve\u00adlo\u00adp\u00adment of high-per\u00adfor\u00admance, ener\u00adgy effi\u00adci\u00adent hete\u00adro\u00adge\u00adneous com\u00adpu\u00adting sys\u00adtems. The \u201cBoltz\u00admann Initia\u00adti\u00adve\u201d lever\u00ada\u00adges HSA<\/span>\u2019s abili\u00adty to harness both cen\u00adtral pro\u00adces\u00adsing units (CPU<\/span>) and AMD<\/span> Fire\u00adPro\u2122 gra\u00adphics pro\u00adces\u00adsing units (GPU<\/span>) for maxi\u00admum com\u00adpu\u00adte effi\u00adci\u00aden\u00adcy through soft\u00adware. The first results of the initia\u00adti\u00adve are fea\u00adtured this week at SC15<\/span><\/a> and include the Hete\u00adro\u00adge\u00adneous Com\u00adpu\u00adte Com\u00adpi\u00adler (HCC<\/span>); a head\u00adless Linux\u00ae dri\u00adver and HSA<\/span> run\u00adtime infra\u00adstruc\u00adtu\u00adre for clus\u00adter-class, High Per\u00adfor\u00admance Com\u00adpu\u00adting (HPC<\/span>); and the Hete\u00adro\u00adge\u00adneous-com\u00adpu\u00adte Inter\u00adface for Por\u00adta\u00adbi\u00adli\u00adty (HIP<\/span>) tool for port\u00ading CUDA-based appli\u00adca\u00adti\u00adons to a com\u00admon C++ pro\u00adgramming model. The tools are desi\u00adgned to dri\u00adve appli\u00adca\u00adti\u00adon per\u00adfor\u00admance across mar\u00adkets ran\u00adging from machi\u00adne lear\u00adning to mole\u00adcu\u00adlar dyna\u00admics, and from oil and gas to visu\u00adal effects and com\u00adpu\u00adter-gene\u00adra\u00adted imaging.<\/p>\n

\u201c<\/span>AMD<\/span>\u2019s Hete\u00adro\u00adge\u00adneous-com\u00adpu\u00adte Inter\u00adface for Por\u00adta\u00adbi\u00adli\u00adty enables performance
\npor\u00adta\u00adbi\u00adli\u00adty for the HPC<\/span> com\u00admu\u00adni\u00adty. The abili\u00adty to take code that was writ\u00adten for one archi\u00adtec\u00adtu\u00adre and trans\u00adfer it to ano\u00adther archi\u00adtec\u00adtu\u00adre wit\u00adhout a nega\u00adti\u00adve impact on per\u00adfor\u00admance is extre\u00adme\u00adly powerful,\u201d said Jim Bel\u00adak, co-lead of the U.S. Depart\u00adment of Energy\u2019s Exas\u00adca\u00adle Co-design Cen\u00adter in Extre\u00adme Mate\u00adri\u00adals and seni\u00ador com\u00adpu\u00adta\u00adtio\u00adnal mate\u00adri\u00adals sci\u00aden\u00adtist at Law\u00adrence Liver\u00admo\u00adre Natio\u00adnal Labo\u00adra\u00adto\u00adry. \u201cThe work AMD<\/span> is doing to pro\u00addu\u00adce a high-per\u00adfor\u00admance com\u00adpi\u00adler that sits below high-level pro\u00adgramming models enables rese\u00adar\u00adchers to con\u00adcen\u00adtra\u00adte on sol\u00adving pro\u00adblems and publi\u00adshing ground\u00adbrea\u00adking rese\u00adarch rather than worry\u00ading about hard\u00adware-spe\u00adci\u00adfic optimizations.\u201d<\/p>\n

New Com\u00adpi\u00adler for Hete\u00adro\u00adge\u00adneous Computing <\/strong><\/p>\n

The pro\u00admi\u00adse of com\u00adbi\u00adning mul\u00adti-core, seri\u00adal pro\u00adces\u00adsing CPUs with par\u00adal\u00adlel-pro\u00adces\u00adsing GPUs to maxi\u00admi\u00adze com\u00adpu\u00adte effi\u00adci\u00aden\u00adcy is alre\u00ada\u00addy being seen in the indus\u00adtry, as dri\u00adven by the Hete\u00adro\u00adge\u00adneous Sys\u00adtems Archi\u00adtec\u00adtu\u00adre (HSA<\/span>) Foun\u00adda\u00adti\u00adon<\/a> that counts AMD<\/span> as a foun\u00adding mem\u00adber. One of the goals for HSA<\/span> is easing the deve\u00adlo\u00adp\u00adment of par\u00adal\u00adlel appli\u00adca\u00adti\u00adons through use of hig\u00adher level lan\u00adguages. The new AMD<\/span> \u201cBoltz\u00admann Initia\u00adti\u00adve\u201d suite includes an HCC<\/span> com\u00adpi\u00adler for C++ deve\u00adlo\u00adp\u00adment, great\u00adly expan\u00adding the field of pro\u00adgramm\u00aders who can levera\u00adge HSA<\/span>. The new HCC<\/span> C++ com\u00adpi\u00adler<\/a> is a key tool in enab\u00adling deve\u00adlo\u00adpers to easi\u00adly and effi\u00adci\u00adent\u00adly app\u00adly the hard\u00adware resour\u00adces in hete\u00adro\u00adge\u00adneous sys\u00adtems. The com\u00adpi\u00adler offers more sim\u00adpli\u00adfied deve\u00adlo\u00adp\u00adment via sin\u00adgle source exe\u00adcu\u00adti\u00adon, with both the CPU<\/span> and GPU<\/span> code in the same file. The com\u00adpi\u00adler auto\u00adma\u00adtes the pla\u00adce\u00adment code that exe\u00adcu\u00adtes on both pro\u00adces\u00adsing ele\u00adments for maxi\u00admum exe\u00adcu\u00adti\u00adon efficiency.<\/p>\n

\u201c<\/span>Just as our cus\u00adto\u00admers are exci\u00adted about our hard\u00adware inno\u00adva\u00adti\u00adon, inclu\u00adding the intro\u00adduc\u00adtion of the first GPU<\/span> with High Band\u00adwidth Memo\u00adry this year and our new x86 core archi\u00adtec\u00adtu\u00adre coming next year, our inno\u00adva\u00adtions in soft\u00adware deve\u00adlo\u00adp\u00adment are equal\u00adly as important to them,\u201d said Mark Paper\u00admas\u00adter, seni\u00ador vice pre\u00adsi\u00addent and chief tech\u00adno\u00adlo\u00adgy offi\u00adcer, AMD<\/span>. \u201cThe chall\u00adenge has always been to unlock the hardware\u2019s capa\u00adbi\u00adli\u00adties and make them easi\u00adly acces\u00adsi\u00adble to deve\u00adlo\u00adpers working to sol\u00adve dif\u00adfi\u00adcult pro\u00adblems. AMD<\/span>\u2019s newest offe\u00adring pro\u00advi\u00addes the keys to more rea\u00addi\u00adly access our par\u00adal\u00adlel com\u00adpu\u00adting engi\u00adnes \u2013 both mul\u00adti\u00adco\u00adre CPUs and GPUs \u2013 and to make the\u00adse bene\u00adfits available to the main\u00adstream of deve\u00adlo\u00adpers across a broad spec\u00adtrum of com\u00adpu\u00adting plat\u00adforms, from embedded to supercomputing.\u201d<\/p>\n

Linux Dri\u00adver and Run\u00adtime Focu\u00adsed on the Needs of HPC<\/span> Clus\u00adter-Class Computing <\/strong><\/p>\n

To com\u00adple\u00adment the new com\u00adpi\u00adla\u00adti\u00adon tools, AMD<\/span> has deve\u00adlo\u00adped a new HPC-focu\u00adsed dri\u00adver and sys\u00adtem run\u00adtime. This new head\u00adless Linux dri\u00adver brings key capa\u00adbi\u00adli\u00adties to address core high-per\u00adfor\u00admance com\u00adpu\u00adting needs, inclu\u00adding low laten\u00adcy com\u00adpu\u00adte dis\u00adpatch and PCIe\u00ae data trans\u00adfers; peer-to-peer GPU<\/span> sup\u00adport; Remo\u00adte Direct Memo\u00adry Access (RDMA<\/span>) from Infi\u00adni\u00adBand\u2122 that inter\u00adcon\u00adnects direct\u00adly to GPU<\/span> memo\u00adry; and Lar\u00adge Sin\u00adgle Memo\u00adry Allo\u00adca\u00adti\u00adon support.<\/p>\n

HIP-ify\u00ading CUDA<\/span> Appli\u00adca\u00adti\u00adon to Run on AMD<\/span> GPUs <\/strong><\/p>\n

To bring appli\u00adca\u00adti\u00adons writ\u00adten for CUDA<\/span> onto AMD<\/span> plat\u00adforms, AMD<\/span> announ\u00adces the new HIP<\/span> tool. AMD<\/span> test\u00ading shows that in many cases 90 per\u00adcent or more of CUDA<\/span> code can be auto\u00adma\u00adti\u00adcal\u00adly con\u00adver\u00adted into C++ by HIP<\/span> with the final 10 per\u00adcent con\u00adver\u00adted manu\u00adal\u00adly in the wide\u00adly popu\u00adlar C++ lan\u00adguage. This great\u00adly expands the instal\u00adled hard\u00adware base available to run what were form\u00ader\u00adly exclu\u00adsi\u00adve\u00adly CUDA-based appli\u00adca\u00adti\u00adons. At SC15<\/span>, AMD<\/span> is demons\u00adt\u00adra\u00adting the poten\u00adti\u00adal for HIP<\/span>, run\u00adning the CUDA-gene\u00adra\u00adted Rodi\u00adnia bench\u00admark suite on AMD<\/span> GPUs.<\/p>\n

Avai\u00adla\u00adbi\u00adli\u00adty<\/strong><\/p>\n

An ear\u00adly access pro\u00adgram for the \u201cBoltz\u00admann Initia\u00adti\u00adve\u201d tools is plan\u00adned for Q1<\/span> 2016.<\/p>\n

Other News at SC15<\/span><\/strong><\/p>\n

AMD<\/span> con\u00adti\u00adnues to sup\u00adport the open stan\u00addard Open\u00adCL\u2122 pro\u00adgramming lan\u00adguage, giving deve\u00adlo\u00adpers low-level GPU<\/span> pro\u00adgramming and opti\u00admiza\u00adti\u00adon capa\u00adbi\u00adli\u00adties. AMD<\/span> has expan\u00added its GPU<\/span> com\u00adpu\u00adte libra\u00adri\u00ades, allo\u00adwing deve\u00adlo\u00adpers to easi\u00adly acce\u00adle\u00adra\u00adte com\u00adpu\u00adte-inten\u00adsi\u00adve func\u00adtions by lever\u00adaging GPU<\/span> com\u00adpu\u00adte. The addi\u00adti\u00adon of HcBLAS and HcFFT builds on work done<\/a> with clB\u00adLAS, clFFT, clS\u00adPAR\u00adSE and clRNG for Open\u00adCL. The libra\u00adri\u00ades include func\u00adtions that are wide\u00adly used in sci\u00aden\u00adti\u00adfic com\u00adpu\u00adta\u00adti\u00adon and data science.<\/p>\n

Indus\u00adtry Sup\u00adport for AMD<\/span><\/strong><\/p>\n

Sup\u00adport con\u00adti\u00adnues to grow for AMD<\/span>\u2019s stan\u00addards-based approach to HPC<\/span>. AMD<\/span>\u2019s invest\u00adment in hard\u00adware and soft\u00adware deve\u00adlo\u00adp\u00adment has resul\u00adted in a gro\u00adwing num\u00adber of sys\u00adtem and solu\u00adti\u00adon ven\u00addors that ful\u00adly sup\u00adport AMD<\/span> Fire\u00adPro\u2122 in enter\u00adpri\u00adse envi\u00adron\u00adments, inclu\u00adding ASRock, ASUS<\/span>, Cir\u00adras\u00adca\u00adle, Cubix, Dell, Giga\u00adbyte, HP<\/span>, One Stop Sys\u00adtems and Supermicro.<\/p>\n

Demons\u00adtra\u00adti\u00adons at SC15<\/span><\/strong><\/p>\n

At SC15<\/span> in booth 727, AMD<\/span> will showcase:<\/p>\n