{"id":960,"date":"2016-12-20T01:59:35","date_gmt":"2016-12-19T22:59:35","guid":{"rendered":"http:\/\/www.teknolojidergisi.net\/?p=960"},"modified":"2016-12-20T01:59:35","modified_gmt":"2016-12-19T22:59:35","slug":"amd-instinct-duyuruldu","status":"publish","type":"post","link":"https:\/\/www.teknolojidergisi.net\/?p=960","title":{"rendered":"AMD Instinct Duyuruldu"},"content":{"rendered":"<p class=\"p1\"><span class=\"s1\">AMD performans, verimlilik ve derin \u00f6\u011frenme i\u015f y\u00fckleri, uygulama kolayl\u0131\u011f\u0131n\u0131 art\u0131rmak i\u00e7in tasarlanm\u0131\u015f yeni donan\u0131m ve a\u00e7\u0131k-kaynak kodlu yaz\u0131l\u0131m paketleri arac\u0131l\u0131\u011f\u0131yla sunucu bilgi i\u015flemde yapay zek\u00e2 \u00e7a\u011f\u0131n\u0131 h\u0131zland\u0131rma stratejisini duyurdu. Radeon Instinct h\u0131zland\u0131r\u0131c\u0131lar, kurulu\u015flara derin \u00f6\u011frenme, \u00e7\u0131kar\u0131m ve e\u011fitimi i\u00e7in g\u00fc\u00e7l\u00fc GPU-tabanl\u0131 \u00e7\u00f6z\u00fcmler sunuyor.<\/span><\/p>\n<p class=\"p1\"><!--more--><\/p>\n<p class=\"p1\"><span class=\"s1\">AMD yeni donan\u0131m sunumlar\u0131n\u0131n yan\u0131 s\u0131ra, y\u00fcksek-performansl\u0131 yapay zek\u00e2 uygulamalar\u0131na olanak sa\u011flamak amac\u0131yla GPU h\u0131zland\u0131r\u0131c\u0131lar i\u00e7in \u00fccretsiz, a\u00e7\u0131k-kaynak kodlu bir k\u00fct\u00fcphane ve bir sonraki makine zek\u00e2s\u0131 i\u015f y\u00fckleri evriminin zeminini haz\u0131rlamak i\u00e7in ROCm yaz\u0131l\u0131m\u0131 \u00fczerinde optimize edilmi\u015f yeni derin \u00f6\u011frenme \u00e7er\u00e7evesini de a\u00e7\u0131klad\u0131.<\/span><\/p>\n<p class=\"p1\"><span class=\"s1\">Ucuz y\u00fcksek kapasiteli depolama, sensor-odakl\u0131 veri bollu\u011fu ve kullan\u0131c\u0131 taraf\u0131ndan geli\u015ftirilen i\u00e7eriklerdeki katlanarak b\u00fcy\u00fcme k\u00fcresel olarak eksabaytlarca veri olu\u015fturuyor. Y\u00fcksek performansl\u0131 GPU\u2019lara e\u015flenen yapay zeka algoritmalar\u0131ndaki son geli\u015fmeler, veri i\u015fleme ve anlamada \u00fcstel bir h\u0131zlanma sa\u011fl\u0131yor ve ger\u00e7ek zamana yak\u0131n \u00f6ng\u00f6r\u00fcler olu\u015fturmaya f\u0131rsat tan\u0131yor. Yapay zek\u00e2 a\u00e7\u0131k yaz\u0131l\u0131m ekosistemi i\u00e7in bir tasar\u0131m olan Radeon Instinct, \u00e7\u0131kar\u0131m \u00f6ng\u00f6r\u00fclerinin ve algoritma e\u011fitiminin h\u0131zland\u0131r\u0131lmas\u0131na yard\u0131mc\u0131 oluyor.<\/span><\/p>\n<p class=\"p1\"><span class=\"s1\">Radeon Instinct h\u0131zland\u0131r\u0131c\u0131lar; pasif so\u011futma, SR-IOV (Tekil K\u00f6k G\/\u00c7 Sanalla\u015ft\u0131rma) end\u00fcstri standard\u0131 ile uyumlu AMD MultiGPU (MxGPU) donan\u0131m sanalla\u015ft\u0131rma teknolojisi ve \u00e7oklu-GPU u\u00e7tan-uca destek i\u00e7in B\u00fcy\u00fck Temel Adres Yazmac\u0131 (BAR) ile 64-bit PCIe adresleme sunuyor.<\/span><\/p>\n<p class=\"p1\"><span class=\"s1\">Radeon Instinct h\u0131zland\u0131r\u0131c\u0131lar, geni\u015f bir yapay zek\u00e2 uygulama yelpazesine hitap etmek i\u00e7in tasarland\u0131:<\/span><\/p>\n<ul class=\"ul1\">\n<li class=\"li1\"><span class=\"s1\">Geni\u015f bir kitle taraf\u0131ndan kabul g\u00f6ren Polaris GPU mimarisine dayal\u0131 Radeon Instinct MI6 h\u0131zland\u0131r\u0131c\u0131, 150W devre kart\u0131 g\u00fcc\u00fc ve 16GB GPU bellekte 5.7 TFLOPS azami FP16 performans\u0131 ile i\u015f\/saniye\/joule i\u00e7in optimize edilmi\u015f pasif so\u011futmal\u0131bir \u00e7\u0131kar\u0131m h\u0131zland\u0131r\u0131c\u0131s\u0131d\u0131r.<\/span><\/li>\n<li class=\"li1\"><span class=\"s1\">Y\u00fcksek performansl\u0131, enerji-verimli \u201cFiji\u201d Nano GPU mimarisine dayal\u0131 Radeon Instinct MI8 h\u0131zland\u0131r\u0131c\u0131, 175W\u2019dan daha d\u00fc\u015f\u00fck devre kart\u0131 g\u00fcc\u00fc ve 4GB Geni\u015f Bant Bellekte (HBM) 8.2 TFLOPS azami FP16 performans\u0131 ile k\u00fc\u00e7\u00fck form fakt\u00f6r\u00fcne sahip bir HPC ve \u00e7\u0131kar\u0131m h\u0131zland\u0131r\u0131c\u0131s\u0131d\u0131r.<\/span><\/li>\n<li class=\"li1\"><span class=\"s1\">Radeon Instinct MI25 h\u0131zland\u0131r\u0131c\u0131 AMD\u2019nin gelecek nesil y\u00fcksek performansl\u0131 Vega GPU mimarisini kullanacak olup, derin \u00f6\u011frenme e\u011fitimi i\u00e7in tasarlanm\u0131\u015f ve \u00e7\u00f6z\u00fcm s\u00fcresi i\u00e7in optimize edildi.<\/span><\/li>\n<\/ul>\n<p class=\"p1\"><span class=\"s1\">Radeon Instinct donan\u0131m\u0131 g\u00fcc\u00fcn\u00fc \u00e7e\u015fitli a\u00e7\u0131k-kaynak kodlu \u00e7\u00f6z\u00fcmlerinden al\u0131yor:<\/span><\/p>\n<ul class=\"ul1\">\n<li class=\"li1\"><span class=\"s1\">MIOpen GPU-h\u0131zland\u0131r\u0131lm\u0131\u015f k\u00fct\u00fcphane: Y\u00fcksek performansl\u0131 yapay zek\u00e2 uygulamalar\u0131n\u0131n \u00e7\u00f6z\u00fcm\u00fcne yard\u0131mc\u0131 olmak \u00fczere, \u00fccretsiz, a\u00e7\u0131k-kaynakl\u0131 MIOpen GPU-h\u0131zland\u0131r\u0131lm\u0131\u015f k\u00fct\u00fcphanenin evri\u015fim, kuyruklama, aktivasyon fonksiyonlar\u0131, normalizasyon ve tens\u00f6r format\u0131 gibi standart rutinler i\u00e7in GPU-ayarl\u0131 uygulamalar sa\u011flamak \u00fczere 2017 Q1\u2019de pazara s\u00fcr\u00fclmesi planlan\u0131yor.<\/span><\/li>\n<li class=\"li1\"><span class=\"s1\">ROCm derin \u00f6\u011frenme \u00e7er\u00e7evesi: ROCm platformu \u015fimdi Caffe, Torch 7 ve TensorFlow* dahil olmak \u00fczere pop\u00fcler derin \u00f6\u011frenme \u00e7er\u00e7eveleri i\u00e7in de optimize edilmi\u015f olup, programc\u0131lar\u0131n ROCm\u2019in zengin entegrasyonlar\u0131 arac\u0131l\u0131\u011f\u0131yla d\u00fc\u015f\u00fck-seviye performans iyile\u015ftirme yerine sinir a\u011flar\u0131n\u0131n e\u011fitimi \u00fczerine odaklanmalar\u0131na olanak sa\u011flamaktad\u0131r. ROCm, do\u011frusal cebir ve tens\u00f6rler i\u00e7in alana-\u00f6zel derleyicilerin yan\u0131 s\u0131ra a\u00e7\u0131k derleyiciler ve dil \u00e7al\u0131\u015fma zaman\u0131 ile yapay zeka problem setlerinin gelecek evriminin temeli olarak hizmet vermek \u00fczere tasarlanm\u0131\u015ft\u0131r.<\/span><\/li>\n<\/ul>\n<p class=\"p1\"><span class=\"s1\">AMD, yar\u0131n\u0131n yapay zek\u00e2 uygulamalar\u0131n\u0131n performans\u0131n\u0131 art\u0131rmak i\u00e7in g\u00fcn\u00fcm\u00fcz\u00fcn PCIe Gen3 standartlar\u0131n\u0131n \u00f6tesine ge\u00e7en ba\u011flant\u0131 teknoloji geli\u015ftirmelerine de yat\u0131r\u0131m yap\u0131yor. AMD, x86, OpenPOWER ve ARM AArch64 de dahi olmak \u00fczere geni\u015f ekosistem sunucu CPU mimarileri yelpazesini destekleyen \u00e7e\u015fitli a\u00e7\u0131k y\u00fcksek-performansl\u0131 I\/O standartlar\u0131 \u00fczerinde i\u015fbirli\u011fi ger\u00e7ekle\u015ftiriyor.<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"<p>AMD performans, verimlilik ve derin \u00f6\u011frenme i\u015f y\u00fckleri, uygulama kolayl\u0131\u011f\u0131n\u0131 art\u0131rmak i\u00e7in tasarlanm\u0131\u015f yeni donan\u0131m ve a\u00e7\u0131k-kaynak kodlu yaz\u0131l\u0131m paketleri arac\u0131l\u0131\u011f\u0131yla sunucu bilgi i\u015flemde yapay zek\u00e2 \u00e7a\u011f\u0131n\u0131 h\u0131zland\u0131rma stratejisini duyurdu. Radeon Instinct h\u0131zland\u0131r\u0131c\u0131lar, kurulu\u015flara derin \u00f6\u011frenme, \u00e7\u0131kar\u0131m ve e\u011fitimi i\u00e7in g\u00fc\u00e7l\u00fc GPU-tabanl\u0131 \u00e7\u00f6z\u00fcmler sunuyor.<\/p>\n","protected":false},"author":3,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[2],"tags":[],"class_list":["post-960","post","type-post","status-publish","format-standard","hentry","category-donanim"],"_links":{"self":[{"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/posts\/960"}],"collection":[{"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=960"}],"version-history":[{"count":1,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/posts\/960\/revisions"}],"predecessor-version":[{"id":961,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=\/wp\/v2\/posts\/960\/revisions\/961"}],"wp:attachment":[{"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=960"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=960"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.teknolojidergisi.net\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=960"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}