{"id":67898,"date":"2023-04-06T16:24:31","date_gmt":"2023-04-06T14:24:31","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=67898"},"modified":"2023-04-06T16:24:31","modified_gmt":"2023-04-06T14:24:31","slug":"amd-launches-first-5nm-asic-based-media-accelerator-card-to-power-new-era-of-interactive-media-services-at-scale","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/67898-amd-launches-first-5nm-asic-based-media-accelerator-card-to-power-new-era-of-interactive-media-services-at-scale\/","title":{"rendered":"AMD<\/span> Launches First 5nm ASIC-based Media Accelerator Card to Power New Era of Interactive Media Services at Scale"},"content":{"rendered":"

Purpose-built video processing architecture featuring an AV1<\/span> accelerated pipeline delivers 32x 1080p streams per card with AI-enabled video quality optimization<\/h3>\n

SANTA<\/span> CLARA<\/span>, Calif., April 06, 2023 (GLOBE<\/span> NEWSWIRE<\/span>) \u2014 AMD<\/span><\/a> (NASDAQ<\/span>: AMD<\/span>) today announ\u00adced the AMD<\/span> Alveo\u2122 MA35D<\/span> media acce\u00adle\u00adra\u00adtor fea\u00adturing two 5nm, ASIC-based video pro\u00adces\u00adsing units (VPUs) sup\u00adport\u00ading the AV1<\/span> com\u00adpres\u00adsi\u00adon stan\u00addard and pur\u00adpo\u00adse-built to power a new era of live inter\u00adac\u00adti\u00adve strea\u00adming ser\u00advices at sca\u00adle. With over 70% of the glo\u00adbal video mar\u00adket being domi\u00adna\u00adted by live con\u00adtent1<\/sup>, a new class of low-laten\u00adcy, high-volu\u00adme inter\u00adac\u00adti\u00adve strea\u00adming appli\u00adca\u00adti\u00adons are emer\u00adging such as watch par\u00adties, live shop\u00adping, online auc\u00adtions, and social streaming.<\/p>\n

The Alveo MA35D<\/span> media acce\u00adle\u00adra\u00adtor deli\u00advers the high chan\u00adnel den\u00adsi\u00adty, with up to 32x 1080p60 streams per card, power effi\u00adci\u00aden\u00adcy and ultra-low-laten\u00adcy per\u00adfor\u00admance cri\u00adti\u00adcal to redu\u00adcing the sky\u00adro\u00adcke\u00adting infra\u00adstruc\u00adtu\u00adre cos\u00adts now requi\u00adred for sca\u00adling such com\u00adpu\u00adte inten\u00adsi\u00adve con\u00adtent deli\u00advery. Com\u00adpared to the pre\u00advious gene\u00adra\u00adti\u00adon Alveo U30<\/span> media acce\u00adle\u00adra\u00adtor, the Alveo MA35D<\/span> deli\u00advers up to 4x hig\u00adher chan\u00adnel den\u00adsi\u00adty2<\/sup>, 4x max lower laten\u00adcy in 4K<\/span>3<\/sup> and 1.8x grea\u00adter com\u00adpres\u00adsi\u00adon effi\u00adci\u00aden\u00adcy4<\/sup> to achie\u00adve the same VMAF<\/span> score \u2013 a com\u00admon video qua\u00adli\u00adty metric. <\/p>\n

\u201c<\/span>We work\u00aded clo\u00adse\u00adly with our cus\u00adto\u00admers and part\u00adners to under\u00adstand not just their tech\u00adni\u00adcal requi\u00adre\u00adments, but their infra\u00adstruc\u00adtu\u00adre chal\u00adlenges in deploy\u00ading high-volu\u00adme, inter\u00adac\u00adti\u00adve strea\u00adming ser\u00advices pro\u00adfi\u00adta\u00adb\u00adly,\u201d said Dan Gib\u00adbons, gene\u00adral mana\u00adger of AECG<\/span> Data Cen\u00adter Group, AMD<\/span>. \u201cWe deve\u00adlo\u00adped the Alveo MA35D<\/span> with an ASIC<\/span> archi\u00adtec\u00adtu\u00adre tail\u00ado\u00adred to meet the bespo\u00adke needs of the\u00adse pro\u00advi\u00adders to redu\u00adce both capi\u00adtal and ope\u00adra\u00adting expen\u00adses for deli\u00adve\u00adring immersi\u00adve expe\u00adri\u00aden\u00adces to their users and con\u00adtent crea\u00adtors at scale.\u201d<\/p>\n

Pur\u00adpo\u00adse-Built Video Pro\u00adces\u00adsing Unit<\/strong>
\nThe Alveo MA35D<\/span> uti\u00adli\u00adzes a pur\u00adpo\u00adse-built VPU<\/span> to acce\u00adle\u00adra\u00adte the enti\u00adre video pipe\u00adline. By per\u00adforming all video pro\u00adces\u00adsing func\u00adtions on the VPU<\/span>, data move\u00adment bet\u00adween the CPU<\/span> and acce\u00adle\u00adra\u00adtor is mini\u00admi\u00adzed, redu\u00adcing over\u00adall laten\u00adcy and maxi\u00admi\u00adzing chan\u00adnel den\u00adsi\u00adty with up to 32x 1080p60, 8x 4Kp60, or 4x 8Kp30 streams per card. The plat\u00adform pro\u00advi\u00addes ultra-low laten\u00adcy sup\u00adport for the main\u00adstream H.264 and H.265 codecs and fea\u00adtures next-gene\u00adra\u00adti\u00adon AV1<\/span> trans\u00adco\u00adder engi\u00adnes deli\u00adve\u00adring up to a 52% reduc\u00adtion in bit\u00adra\u00adte for band\u00adwidth savings ver\u00adsus a com\u00adpa\u00adra\u00adble soft\u00adware imple\u00admen\u00adta\u00adti\u00adon5<\/sup>.<\/p>\n

\u201c<\/span>AMD<\/span>\u2019s announce\u00adment of the new Alveo MA35D<\/span> add-in card is an exci\u00adting advance\u00adment of video acce\u00adle\u00adra\u00adti\u00adon for data cen\u00adters and is an important step in buil\u00adding out a ful\u00adly-fled\u00adged eco\u00adsys\u00adtem to sup\u00adport royal\u00adty-free, high-defi\u00adni\u00adti\u00adon video devices, pro\u00adducts, and ser\u00advices,\u201d said Matt Frost, Alli\u00adance for Open Media Chair. \u201cLive strea\u00adming pro\u00advi\u00adders are loo\u00adking for hig\u00adher den\u00adsi\u00adty, lower power, lower laten\u00adcy AV1<\/span> solu\u00adti\u00adons and by addres\u00adsing the\u00adse, Alli\u00adance mem\u00adbers such as AMD<\/span> are hel\u00adping faci\u00adli\u00adta\u00adte AV1<\/span> deploy\u00adment and over\u00adall adoption.\u201d <\/p>\n

AI-Enab\u00adled, Intel\u00adli\u00adgent Video Pipeline<\/strong>
\nThe acce\u00adle\u00adra\u00adtor fea\u00adtures an inte\u00adgra\u00adted AI<\/span> pro\u00adces\u00adsor and dedi\u00adca\u00adted video qua\u00adli\u00adty engi\u00adnes desi\u00adgned to impro\u00adve the qua\u00adli\u00adty of expe\u00adri\u00adence at redu\u00adced band\u00adwidth. The AI<\/span> pro\u00adces\u00adsor eva\u00adlua\u00adtes con\u00adtent, frame-by-frame, and dyna\u00admi\u00adcal\u00adly adjus\u00adts enco\u00adder set\u00adtings to impro\u00adve per\u00adcei\u00adved visu\u00adal qua\u00adli\u00adty while mini\u00admi\u00adzing bit\u00adra\u00adte. Opti\u00admiza\u00adti\u00adon tech\u00adni\u00adques include regi\u00adon-of-inte\u00adrest (ROI<\/span>) enco\u00adding for text and face reso\u00adlu\u00adti\u00adon, arti\u00adfact detec\u00adtion to cor\u00adrect sce\u00adnes with high levels of moti\u00adon and com\u00adple\u00adxi\u00adty, and con\u00adtent-awa\u00adre enco\u00adding for pre\u00addic\u00adti\u00adve insights for bit\u00adra\u00adte optimization.<\/p>\n

 Cost-Effec\u00adtively Sca\u00adle Inter\u00adac\u00adti\u00adve Media<\/strong>
\nSca\u00adling high-volu\u00adme strea\u00adming ser\u00advices requi\u00adres maxi\u00admi\u00adzing the num\u00adber of chan\u00adnels per ser\u00adver while mini\u00admi\u00adzing power and band\u00adwidth-per-stream. By deli\u00adve\u00adring up to 32x 1080p60 streams per card at 1 watt per stream6<\/sup>, a 1U<\/span> rack ser\u00adver equip\u00adped with 8 cards deli\u00advers up to 256 chan\u00adnels to maxi\u00admi\u00adze the num\u00adber of streams per ser\u00adver, rack or data center.<\/p>\n

Soft\u00adware Dev Kit and Pro\u00adduct Availability<\/strong>
\nThe plat\u00adform is acces\u00adsi\u00adble with the AMD<\/span> Media Acce\u00adle\u00adra\u00adti\u00adon soft\u00adware deve\u00adlo\u00adp\u00adment kit (SDK<\/span>), sup\u00adport\u00ading the wide\u00adly used FFmpeg and Gstrea\u00admer video frame\u00adworks for ease of development.<\/p>\n

Alveo MA35D<\/span> media acce\u00adle\u00adra\u00adtors are sam\u00adpling now with pro\u00adduc\u00adtion ship\u00adments expec\u00adted in Q3<\/span>. To acce\u00adle\u00adra\u00adte deve\u00adlo\u00adp\u00adment, an Ear\u00adly Access Pro\u00adgram is available to qua\u00adli\u00adfied cus\u00adto\u00admers with com\u00adpre\u00adhen\u00adsi\u00adve docu\u00admen\u00adta\u00adti\u00adon and soft\u00adware tools for archi\u00adtec\u00adtu\u00adral exploration.<\/p>\n

Sup\u00adport\u00ading Resources<\/strong><\/p>\n