{"id":59863,"date":"2020-11-16T15:00:13","date_gmt":"2020-11-16T14:00:13","guid":{"rendered":"https:\/\/www.planet3dnow.de\/cms\/?p=59863"},"modified":"2020-11-16T15:01:30","modified_gmt":"2020-11-16T14:01:30","slug":"amd-stellt-instinct-m100-vor-und-will-nvidia-damit-angreifen","status":"publish","type":"post","link":"https:\/\/www.planet3dnow.de\/cms\/59863-amd-stellt-instinct-m100-vor-und-will-nvidia-damit-angreifen\/","title":{"rendered":"AMD<\/span> stellt Instinct MI100<\/span> vor und will Nvidia damit angreifen"},"content":{"rendered":"

Mit der Instinct MI100<\/span> stellt AMD<\/span> einen Gra\u00adfik\u00adbe\u00adschleu\u00adni\u00adger f\u00fcr das High Per\u00adfor\u00admance Com\u00adpu\u00adting vor, der AMD<\/span> nach eige\u00adnen Aus\u00adsa\u00adgen an die Spit\u00adze in die\u00adsem Bereich kata\u00adpul\u00adtie\u00adren soll. Neben den rei\u00adnen Per\u00adfor\u00admance\u00adver\u00adbes\u00adse\u00adrun\u00adgen soll dies auch durch einen Aus\u00adbau des Soft\u00adware\u00ad\u00f6ko\u00adsys\u00adtems erreicht wer\u00adden. Dazu soll vor allem die neue Ver\u00adsi\u00adon 4.0 der Rade\u00adon Open Com\u00adpu\u00adte Plat\u00adform (ROCm) beitragen.<\/p>\n

<\/a><\/p>\n

Instinct MI100<\/span><\/h3>\n

Die AMD<\/span> Instinct MI100-GPU<\/span> \u2014 dem ein oder ande\u00adrem viel\u00adleicht bereits unter dem Code\u00adna\u00admen \u201cArc\u00adturus\u201d bekannt \u2014 basiert auf der neu\u00aden AMD<\/span> CDNA-Archi\u00adtek\u00adtur und soll laut AMD<\/span> vor allem in Kom\u00adbi\u00adna\u00adti\u00adon mit AMD<\/span> EPYC-Pro\u00adzes\u00adso\u00adren der 2. Gene\u00adra\u00adti\u00adon eine neue Klas\u00adse bei den Beschleu\u00adni\u00adger\u00adkar\u00adten f\u00fcr HPC<\/span> und AI<\/span> dar\u00adstel\u00adlen. Sie besteht aus 120 Com\u00adpu\u00adte Units und 7.680 Streamprozessoren.<\/p>\n

Die Instinct MI100<\/span> bie\u00adtet bis zu 11,5 TFlops bei FP64<\/span> und bis zu 46,1 TFLOPS<\/span> bei FP32-Matrix f\u00fcr AI-<\/span> und maschi\u00adnel\u00adles Ler\u00adnen. Mit einer neu\u00aden AMD<\/span> Matrix Core genann\u00adten Tech\u00adno\u00adlo\u00adgie soll die MI100<\/span> im Ver\u00adgleich zu AMDs Beschleu\u00adni\u00adgern der vor\u00adhe\u00adri\u00adgen Gene\u00adra\u00adti\u00adon eine fast 7\u2011fache Stei\u00adge\u00adrung der theo\u00adre\u00adti\u00adschen FP16-Spit\u00adzen-Gleit\u00adkom\u00adma-Leis\u00adtung f\u00fcr KI-Trai\u00adnings-Workloads erreichen.<\/p>\n

<\/a>
Per\u00adfor\u00admance\u00adver\u00adgleich MI100<\/span> und MI50<\/span> (Quel\u00adle: AMD<\/span>)<\/figcaption><\/figure>\n

Mit Bfloat16 (brain floa\u00adting point with 16 bits) wird nun auch ein neu\u00ades Gleit\u00adkom\u00adma\u00adfor\u00admat unter\u00adst\u00fctzt, wel\u00adches bereits bei Intel in Pro\u00adzes\u00adso\u00adren und FPGAs sowie bei NVI\u00adDI\u00adAs A100<\/span> zum Ein\u00adsatz kommt und beson\u00adders f\u00fcr maschi\u00adnel\u00adles Ler\u00adnen gebraucht wird.<\/p>\n

<\/a>
Auf\u00adbau des AMD<\/span> Instinct MI100<\/span><\/figcaption><\/figure>\n

Angriff auf Nvidias A100<\/span>?<\/h3>\n

Laut AMD<\/span> will man mit der neu\u00aden Beschleu\u00adni\u00adger\u00adkar\u00adte Instinct MI100<\/span> Nvi\u00addi\u00adas Spit\u00adzen\u00adpro\u00addukt A100<\/span> in zahl\u00adrei\u00adchen F\u00e4l\u00adlen \u00fcber\u00adtref\u00adfen, in Berei\u00adchen, in denen dies nicht m\u00f6g\u00adlich ist, will man aller\u00addings bei der Per\u00adfor\u00admance-per-Dol\u00adlar Metrik vor\u00adne liegen.<\/p>\n

 <\/p>\n

<\/a>
Ver\u00adgleich A100<\/span> und Instinct MI100<\/span> (Quel\u00adle: AMD<\/span>)<\/figcaption><\/figure>\n

Ob die\u00adser Angriff gelingt, wird sich erst in der Zukunft zei\u00adgen. Aller\u00addings hat man bei AMD<\/span> bereits zahl\u00adrei\u00adche Super\u00adcom\u00adpu\u00adter ange\u00adk\u00fcn\u00addigt, die auf eine Kom\u00adbi\u00adna\u00adti\u00adon von AMD<\/span> Instinct und Epyc \u2014 also Beschleu\u00adni\u00adger\u00adkar\u00adten und Pro\u00adzes\u00adso\u00adren \u2014 set\u00adzen. An der Spit\u00adze die Exas\u00adcale\u00adsys\u00adte\u00adme Fron\u00adtier und El Capitan.<\/p>\n

 <\/p>\n

Unterst\u00fctzung von den gro\u00dfen Serverherstellern<\/h3>\n

Zum Start pr\u00e4\u00adsen\u00adtiert man dann auch die Unter\u00adst\u00fct\u00adzung von Ser\u00adver\u00adher\u00adstel\u00adlern wie Dell, Giga\u00adbyte, Hew\u00adlett Packard Enter\u00adpri\u00adse (HPE<\/span>) und Supermicro.<\/p>\n

 <\/p>\n

Dell<\/b><\/span><\/span><\/span><\/p>\n

\u201c<\/span>Dell EMC<\/span> PowerEdge ser\u00advers will sup\u00adport the new AMD<\/span> Instinct MI100<\/span>, which will enable fas\u00adter insights from data. This would help our cus\u00adto\u00admers achie\u00adve more robust and effi\u00adci\u00adent HPC<\/span> and AI<\/span> results rapidly,\u201d said Ravi Pen\u00adde\u00adkan\u00adti, seni\u00ador vice pre\u00adsi\u00addent, PowerEdge Ser\u00advers, Dell Tech\u00adno\u00adlo\u00adgies. \u201cAMD<\/span> has been a valued part\u00adner in our sup\u00adport for advan\u00adcing inno\u00adva\u00adti\u00adon in the data cen\u00adter. The high-per\u00adfor\u00admance capa\u00adbi\u00adli\u00adties of AMD<\/span> Instinct acce\u00adle\u00adra\u00adtors are a natu\u00adral fit for our PowerEdge ser\u00adver AI<\/span> &<\/span> HPC<\/span> portfolio.\u201d
\n<\/span><\/span><\/span><\/p>\n

Giga\u00adbyte<\/b><\/span><\/span><\/span><\/p>\n

\u201c<\/span>We\u2019re plea\u00adsed to again work with AMD<\/span> as a stra\u00adte\u00adgic part\u00adner offe\u00adring cus\u00adto\u00admers ser\u00adver hard\u00adware for high per\u00adfor\u00admance com\u00adpu\u00adting,\u201d said Alan Chen, assistant vice pre\u00adsi\u00addent in NCBU<\/span>, GIGABYTE<\/span>. \u201cAMD<\/span> Instinct MI100<\/span> acce\u00adle\u00adra\u00adtors repre\u00adsent the next level of high-per\u00adfor\u00admance com\u00adpu\u00adting in the data cen\u00adter, brin\u00adging grea\u00adter con\u00adnec\u00adti\u00advi\u00adty and data band\u00adwidth for ener\u00adgy rese\u00adarch, mole\u00adcu\u00adlar dyna\u00admics, and deep lear\u00adning trai\u00adning. As a new acce\u00adle\u00adra\u00adtor in the GIGABYTE<\/span> port\u00adfo\u00adlio, our cus\u00adto\u00admers can look to bene\u00adfit from impro\u00adved per\u00adfor\u00admance across a ran\u00adge of sci\u00aden\u00adti\u00adfic and indus\u00adtri\u00adal HPC<\/span> workloads.\u201d<\/span><\/span><\/span><\/p>\n

Hew\u00adlett Packard Enter\u00adpri\u00adse (HPE<\/span>)<\/b><\/span><\/span><\/span><\/p>\n

\u201c<\/span>Cus\u00adto\u00admers use HPE<\/span> Apol\u00adlo sys\u00adtems for pur\u00adpo\u00adse-built capa\u00adbi\u00adli\u00adties and per\u00adfor\u00admance to tack\u00adle a ran\u00adge of com\u00adplex, data-inten\u00adsi\u00adve workloads across high-per\u00adfor\u00admance com\u00adpu\u00adting (HPC<\/span>), deep lear\u00adning and ana\u00adly\u00adtics,\u201d said Bill Man\u00adnel, vice pre\u00adsi\u00addent and gene\u00adral mana\u00adger, HPC<\/span> at HPE<\/span>. \u201cWith the intro\u00adduc\u00adtion of the new HPE<\/span> Apol\u00adlo 6500 Gen10 Plus sys\u00adtem, we are fur\u00adther advan\u00adcing our port\u00adfo\u00adlio to impro\u00adve workload per\u00adfor\u00admance by sup\u00adport\u00ading the new AMD<\/span> Instinct MI100<\/span> acce\u00adle\u00adra\u00adtor, which enables grea\u00adter con\u00adnec\u00adti\u00advi\u00adty and data pro\u00adces\u00adsing, along\u00adside the 2<\/span>nd<\/span><\/sup> Gen AMD<\/span> EPYC<\/span>\u2122 pro\u00adces\u00adsor. We look for\u00adward to con\u00adti\u00adnuing our col\u00adla\u00adbo\u00adra\u00adti\u00adon with AMD<\/span> to expand our offe\u00adrings with its latest CPUs and accelerators.\u201d<\/span><\/span><\/span><\/p>\n

Super\u00admi\u00adcro<\/b><\/span><\/span><\/p>\n

\u201c<\/span>We\u2019re exci\u00adted that AMD<\/span> is making a big impact in high-per\u00adfor\u00admance com\u00adpu\u00adting with AMD<\/span> Instinct MI100<\/span> GPU<\/span> acce\u00adle\u00adra\u00adtors,\u201d said Vik Malya\u00adla, seni\u00ador vice pre\u00adsi\u00addent, field appli\u00adca\u00adti\u00adon engi\u00adnee\u00adring and busi\u00adness deve\u00adlo\u00adp\u00adment, Super\u00admi\u00adcro. \u201cWith the com\u00adbi\u00adna\u00adti\u00adon of the com\u00adpu\u00adte power gai\u00adned with the new CDNA<\/span> archi\u00adtec\u00adtu\u00adre, along with the high memo\u00adry and GPU<\/span> peer-to-peer band\u00adwidth the MI100<\/span> brings, our cus\u00adto\u00admers will get access to gre\u00adat solu\u00adti\u00adons that will meet their acce\u00adle\u00adra\u00adted com\u00adpu\u00adte requi\u00adre\u00adments and <\/span><\/span>cri\u00adti\u00adcal enter\u00adpri\u00adse workloads. The AMD<\/span> Instinct MI100<\/span> will be a gre\u00adat addi\u00adti\u00adon for our mul\u00adti-GPU ser\u00advers and our exten\u00adsi\u00adve port\u00adfo\u00adlio of high-per\u00adfor\u00admance sys\u00adtems and ser\u00adver buil\u00adding block solutions.\u201d<\/span><\/span><\/p>\n

 <\/p>\n

 <\/p>\nngg_shortcode_0_placeholder ","protected":false},"excerpt":{"rendered":"

Mit der Instinct MI100<\/span> stellt AMD<\/span> einen Gra\u00adfik\u00adbe\u00adschleu\u00adni\u00adger f\u00fcr das High Per\u00adfor\u00admance Com\u00adpu\u00adting vor, der AMD<\/span> nach eige\u00adnen Aus\u00adsa\u00adgen an die Spit\u00adze in die\u00adsem Bereich kata\u00adpul\u00adtie\u00adren soll. Neben den rei\u00adnen Per\u00adfor\u00admance\u00adver\u00adbes\u00adse\u00adrun\u00adgen soll dies auch durch einen Aus\u00adbau des Soft\u00adware\u00ad\u00f6ko\u00adsys\u00adtems erreicht wer\u00adden. Dazu soll vor allem die neue Ver\u00adsi\u00adon 4.0 der Rade\u00adon Open Com\u00adpu\u00adte Plat\u00adform (ROCm) bei\u00adtra\u00adgen. (\u2026) Wei\u00adter\u00adle\u00adsen \u00bb<\/a><\/p>\n","protected":false},"author":593,"featured_media":48484,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"wp_typography_post_enhancements_disabled":false,"ngg_post_thumbnail":0,"footnotes":""},"categories":[12,11],"tags":[966,540,2464],"class_list":["post-59863","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-aktuelles","category-news","tag-amd","tag-hpc","tag-instinct-mi100","entry"],"share_on_mastodon":{"url":"","error":""},"_links":{"self":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/59863","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/users\/593"}],"replies":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/comments?post=59863"}],"version-history":[{"count":8,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/59863\/revisions"}],"predecessor-version":[{"id":59874,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/posts\/59863\/revisions\/59874"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/media\/48484"}],"wp:attachment":[{"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/media?parent=59863"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/categories?post=59863"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.planet3dnow.de\/cms\/wp-json\/wp\/v2\/tags?post=59863"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}