AMD Launches ‘Boltzmann Initiative’ to Dramatically Reduce Barriers to GPU Computing on AMD FirePro Graphics

Tools Tar­get an Unpre­ce­den­ted 28 Tera­flops of Pro­ces­sing at Less Than a Kilo­watt by 2016

AUSTIN, Texas — Novem­ber 16, 2015 — Buil­ding on its stra­te­gic invest­ments in hete­ro­ge­neous sys­tem archi­tec­tu­re (HSA), AMD (NASDAQ:AMD) announ­ced a suite of tools desi­gned to ease deve­lo­p­ment of high-per­for­mance, ener­gy effi­ci­ent hete­ro­ge­neous com­pu­ting sys­tems. The “Boltz­mann Initia­ti­ve” lever­a­ges HSA’s abili­ty to harness both cen­tral pro­ces­sing units (CPU) and AMD Fire­Pro™ gra­phics pro­ces­sing units (GPU) for maxi­mum com­pu­te effi­ci­en­cy through soft­ware. The first results of the initia­ti­ve are fea­tured this week at SC15 and include the Hete­ro­ge­neous Com­pu­te Com­pi­ler (HCC); a head­less Linux® dri­ver and HSA run­time infra­struc­tu­re for clus­ter-class, High Per­for­mance Com­pu­ting (HPC); and the Hete­ro­ge­neous-com­pu­te Inter­face for Por­ta­bi­li­ty (HIP) tool for port­ing CUDA-based appli­ca­ti­ons to a com­mon C++ pro­gramming model. The tools are desi­gned to dri­ve appli­ca­ti­on per­for­mance across mar­kets ran­ging from machi­ne lear­ning to mole­cu­lar dyna­mics, and from oil and gas to visu­al effects and com­pu­ter-gene­ra­ted imaging.

AMD’s Hete­ro­ge­neous-com­pu­te Inter­face for Por­ta­bi­li­ty enables performance
por­ta­bi­li­ty for the HPC com­mu­ni­ty. The abili­ty to take code that was writ­ten for one archi­tec­tu­re and trans­fer it to ano­ther archi­tec­tu­re wit­hout a nega­ti­ve impact on per­for­mance is extre­me­ly powerful,” said Jim Bel­ak, co-lead of the U.S. Depart­ment of Energy’s Exas­ca­le Co-design Cen­ter in Extre­me Mate­ri­als and seni­or com­pu­ta­tio­nal mate­ri­als sci­en­tist at Law­rence Liver­mo­re Natio­nal Labo­ra­to­ry. “The work AMD is doing to pro­du­ce a high-per­for­mance com­pi­ler that sits below high-level pro­gramming models enables rese­ar­chers to con­cen­tra­te on sol­ving pro­blems and publi­shing ground­brea­king rese­arch rather than worry­ing about hard­ware-spe­ci­fic optimizations.”

New Com­pi­ler for Hete­ro­ge­neous Computing 

The pro­mi­se of com­bi­ning mul­ti-core, seri­al pro­ces­sing CPUs with par­al­lel-pro­ces­sing GPUs to maxi­mi­ze com­pu­te effi­ci­en­cy is alre­a­dy being seen in the indus­try, as dri­ven by the Hete­ro­ge­neous Sys­tems Archi­tec­tu­re (HSA) Foun­da­ti­on that counts AMD as a foun­ding mem­ber. One of the goals for HSA is easing the deve­lo­p­ment of par­al­lel appli­ca­ti­ons through use of hig­her level lan­guages. The new AMD “Boltz­mann Initia­ti­ve” suite includes an HCC com­pi­ler for C++ deve­lo­p­ment, great­ly expan­ding the field of pro­gramm­ers who can levera­ge HSA. The new HCC C++ com­pi­ler is a key tool in enab­ling deve­lo­pers to easi­ly and effi­ci­ent­ly app­ly the hard­ware resour­ces in hete­ro­ge­neous sys­tems. The com­pi­ler offers more sim­pli­fied deve­lo­p­ment via sin­gle source exe­cu­ti­on, with both the CPU and GPU code in the same file. The com­pi­ler auto­ma­tes the pla­ce­ment code that exe­cu­tes on both pro­ces­sing ele­ments for maxi­mum exe­cu­ti­on efficiency.

Just as our cus­to­mers are exci­ted about our hard­ware inno­va­ti­on, inclu­ding the intro­duc­tion of the first GPU with High Band­width Memo­ry this year and our new x86 core archi­tec­tu­re coming next year, our inno­va­tions in soft­ware deve­lo­p­ment are equal­ly as important to them,” said Mark Paper­mas­ter, seni­or vice pre­si­dent and chief tech­no­lo­gy offi­cer, AMD. “The chall­enge has always been to unlock the hardware’s capa­bi­li­ties and make them easi­ly acces­si­ble to deve­lo­pers working to sol­ve dif­fi­cult pro­blems. AMD’s newest offe­ring pro­vi­des the keys to more rea­di­ly access our par­al­lel com­pu­ting engi­nes – both mul­ti­co­re CPUs and GPUs – and to make the­se bene­fits available to the main­stream of deve­lo­pers across a broad spec­trum of com­pu­ting plat­forms, from embedded to supercomputing.”

Linux Dri­ver and Run­time Focu­sed on the Needs of HPC Clus­ter-Class Computing 

To com­ple­ment the new com­pi­la­ti­on tools, AMD has deve­lo­ped a new HPC-focu­sed dri­ver and sys­tem run­time. This new head­less Linux dri­ver brings key capa­bi­li­ties to address core high-per­for­mance com­pu­ting needs, inclu­ding low laten­cy com­pu­te dis­patch and PCIe® data trans­fers; peer-to-peer GPU sup­port; Remo­te Direct Memo­ry Access (RDMA) from Infi­ni­Band™ that inter­con­nects direct­ly to GPU memo­ry; and Lar­ge Sin­gle Memo­ry Allo­ca­ti­on support.

HIP-ify­ing CUDA Appli­ca­ti­on to Run on AMD GPUs 

To bring appli­ca­ti­ons writ­ten for CUDA onto AMD plat­forms, AMD announ­ces the new HIP tool. AMD test­ing shows that in many cases 90 per­cent or more of CUDA code can be auto­ma­ti­cal­ly con­ver­ted into C++ by HIP with the final 10 per­cent con­ver­ted manu­al­ly in the wide­ly popu­lar C++ lan­guage. This great­ly expands the instal­led hard­ware base available to run what were form­er­ly exclu­si­ve­ly CUDA-based appli­ca­ti­ons. At SC15, AMD is demons­t­ra­ting the poten­ti­al for HIP, run­ning the CUDA-gene­ra­ted Rodi­nia bench­mark suite on AMD GPUs.

Avai­la­bi­li­ty

An ear­ly access pro­gram for the “Boltz­mann Initia­ti­ve” tools is plan­ned for Q1 2016.

Other News at SC15

AMD con­ti­nues to sup­port the open stan­dard Open­CL™ pro­gramming lan­guage, giving deve­lo­pers low-level GPU pro­gramming and opti­miza­ti­on capa­bi­li­ties. AMD has expan­ded its GPU com­pu­te libra­ri­es, allo­wing deve­lo­pers to easi­ly acce­le­ra­te com­pu­te-inten­si­ve func­tions by lever­aging GPU com­pu­te. The addi­ti­on of HcBLAS and HcFFT builds on work done with clB­LAS, clFFT, clS­PAR­SE and clRNG for Open­CL. The libra­ri­es include func­tions that are wide­ly used in sci­en­ti­fic com­pu­ta­ti­on and data science.

Indus­try Sup­port for AMD

Sup­port con­ti­nues to grow for AMD’s stan­dards-based approach to HPC. AMD’s invest­ment in hard­ware and soft­ware deve­lo­p­ment has resul­ted in a gro­wing num­ber of sys­tem and solu­ti­on ven­dors that ful­ly sup­port AMD Fire­Pro™ in enter­pri­se envi­ron­ments, inclu­ding ASRock, ASUS, Cir­ras­ca­le, Cubix, Dell, Giga­byte, HP, One Stop Sys­tems and Supermicro.

Demons­tra­ti­ons at SC15

At SC15 in booth 727, AMD will showcase:

  • The new HSA-com­pli­ant C++ com­pi­ler, full open source HSA pro­gramming suite and HIP port­ing tools.
  • Sys­tems from Dell, HP and Super­mi­cro that include AMD Fire­Pro™ GPUs run­ning demons­tra­ti­ons such as AMD Fire­Ren­der, Aba­qus and TUM Navier-Strokes.
  • AMD’s Opte­ron™ A1100 series ARM pro­ces­sor in SoftIron’s new Enter­pri­se Class Over­dri­ve 3000 sys­tem for deve­lo­pers and Sil­ver Lining System’s ARM-Metal-as-a-Ser­vice (AMAAS) solution.
  • The AMD Fire­Pro™ S9170 ser­ver GPU in action with Dell and Super­mi­cro ser­vers, show­ca­sing a varie­ty of workloads that take advan­ta­ge of the AMD Fire­Pro S9170 card’s 32GB high-speed and high-band­width onboard memo­ry, making it ide­al for acce­le­ra­ting lar­ge data­set computations.
  • AMD engi­neers are sche­du­led to pre­sent two rela­ted tech talks 
    • Tues­day, Novem­ber 17 at 4:15 p.m. Greg Rod­gers, design engi­neer, AMD, “Advan­ces in HSA Soft­ware Stack”
    • Wed­nes­day, Novem­ber 18 at 5:15 p.m. Sri­kanth Gol­la­pu­di, PMTS soft­ware engi­neer, AMD, “Port­ing CUDA appli­ca­ti­ons to C++ with New AMD HIP con­ver­si­on tools”

Sup­port­ing Resources

About AMD

For more than 45 years AMD has dri­ven inno­va­ti­on in high-per­for­mance com­pu­ting, gra­phics, and visua­liza­ti­on tech­no­lo­gies ― the buil­ding blocks for gam­ing, immersi­ve plat­forms, and the dat­a­cen­ter. Hundreds of mil­li­ons of con­su­mers, lea­ding For­tu­ne 500 busi­nesses, and cut­ting-edge sci­en­ti­fic rese­arch faci­li­ties around the world rely on AMD tech­no­lo­gy dai­ly to impro­ve how they live, work, and play. AMD employees around the world are focu­sed on buil­ding gre­at pro­ducts that push the boun­da­ries of what is pos­si­ble. For more infor­ma­ti­on about how AMD is enab­ling today and inspi­ring tomor­row, visit the AMD (NASDAQ: AMD) web­site, blog, Face­book and Twit­ter pages.

—30—

AMD, the AMD Arrow logo, Fire­Pro, Opte­ron, and com­bi­na­ti­ons the­reof, are trade­marks of Advan­ced Micro Devices, Inc. Open­CL is a trade­mark of Apple Inc. used by per­mis­si­on by Khro­nos. Other names are for infor­ma­tio­nal pur­po­ses only and may be trade­marks of their respec­ti­ve owners.