{"id":115715,"date":"2025-12-29T10:40:09","date_gmt":"2025-12-29T10:40:09","guid":{"rendered":"https:\/\/ekamu.net\/?p=115715"},"modified":"2025-12-29T10:40:09","modified_gmt":"2025-12-29T10:40:09","slug":"nvidia-feynman-mimarisinden-ilk-detaylar-paylasildi-lpu-x3d-ve-dahasi","status":"publish","type":"post","link":"https:\/\/ekamu.net\/index.php\/2025\/12\/29\/nvidia-feynman-mimarisinden-ilk-detaylar-paylasildi-lpu-x3d-ve-dahasi\/","title":{"rendered":"Nvidia Feynman mimarisinden ilk detaylar payla\u015f\u0131ld\u0131: LPU, X3D ve dahas\u0131"},"content":{"rendered":"

<\/span> Nvidia’n\u0131n gelecek nesil Feynman GPU mimarisiyle birlikte \u00e7\u0131kar\u0131m odakl\u0131 \u00f6nemli bir de\u011fi\u015fime gidebilece\u011fi konu\u015fuluyor. Aktar\u0131lan bilgilere g\u00f6re \u015firket, 3D yap\u0131ya ek olarak Groq’a ait LPU (Language Processing Unit) birimlerini mimariye entegre etmeyi hedefliyor. Bu yakla\u015f\u0131m\u0131n 2028 sonras\u0131 \u00fcr\u00fcnlerde<\/strong> hayata ge\u00e7irilmesi bekleniyor. <\/figure>\n

2028 y\u0131l\u0131ndan sonra geliyor<\/b><\/p>\n

GPU uzman\u0131 AGF’ye g\u00f6re Nvidia, LPU entegrasyonu i\u00e7in AMD’nin X3D i\u015flemcilerde kulland\u0131\u011f\u0131 yakla\u015f\u0131ma benzer bir yol<\/strong> izleyebilir. Buna g\u00f6re LPU birimleri, ana hesaplama kal\u0131b\u0131ndan ayr\u0131 birer yonga olarak tasarlanacak ve TSMC’nin SoIC hibrit ba\u011flama teknolojisiyle Feynman GPU’lar\u0131n \u00fczerine istiflenecek. Bu sayede, y\u00fcksek bant geni\u015fli\u011fi ve paket d\u0131\u015f\u0131 belle\u011fe k\u0131yasla daha d\u00fc\u015f\u00fck enerji t\u00fcketimi hedefleniyor.<\/p>\n

<\/span> \u015eu an i\u00e7in SRAM’i do\u011frudan monolitik bir kal\u0131p olarak ana GPU’ya entegre etmek, \u00f6l\u00e7eklenebilirlik a\u00e7\u0131s\u0131ndan verimli bulunmuyor. SRAM’in ileri \u00fcretim d\u00fc\u011f\u00fcmlerinde yer kaplamas\u0131, hem maliyetleri art\u0131r\u0131yor hem de \u00fcst seviye silikon alan\u0131n\u0131n verimsiz kullan\u0131lmas\u0131na yol a\u00e7\u0131yor. Bu nedenle Nvidia’n\u0131n, hesaplama birimlerini bar\u0131nd\u0131ran ana Feynman kal\u0131b\u0131n\u0131 1.6 nm A16 s\u00fcrecinde<\/strong> \u00fcretirken, geni\u015f SRAM’e sahip LPU kal\u0131plar\u0131n\u0131 ayr\u0131 olarak konumland\u0131rmas\u0131 daha olas\u0131. <\/figure>\n

TSMC’nin hibrit ba\u011flama teknolojisi burada kritik bir rol \u00fcstleniyor. A16 s\u00fcrecinin sundu\u011fu arka y\u00fcz g\u00fc\u00e7 da\u011f\u0131t\u0131m\u0131 sayesinde, \u00f6n y\u00fczey dikey SRAM ba\u011flant\u0131lar\u0131 i\u00e7in kullan\u0131labiliyor. Bu yap\u0131, d\u00fc\u015f\u00fck gecikmeli veri eri\u015fimi ve \u00e7\u0131kar\u0131m senaryolar\u0131nda daha h\u0131zl\u0131 yan\u0131t s\u00fcreleri sunabilir. Ancak bu yakla\u015f\u0131m\u0131n ciddi m\u00fchendislik zorluklar\u0131 da bulunuyor.<\/p>\n

LPU, X3D ve dahas\u0131<\/b><\/p>\n

Y\u00fcksek hesaplama yo\u011funlu\u011funa sahip bir GPU’nun \u00fczerine ek yongalar istiflemek, termal s\u0131n\u0131rlar\u0131n y\u00f6netimini<\/strong> zorla\u015ft\u0131rabilir. \u00d6zellikle s\u00fcrekli y\u00fcksek i\u015f hacmiyle \u00e7al\u0131\u015fan LPU’lar\u0131n, \u0131s\u0131 ve darbo\u011faz risklerini art\u0131rabilece\u011fi belirtiliyor. Ayr\u0131ca LPU’lar\u0131n sabit y\u00fcr\u00fctme d\u00fczenine sahip olmas\u0131, GPU’lar\u0131n esnek ve paralel yap\u0131s\u0131yla \u00e7eli\u015fen bir yap\u0131 olu\u015fturuyor.<\/p>\n

Yaz\u0131l\u0131m taraf\u0131nda ise CUDA ekosistemi \u00f6nemli bir soru i\u015fareti. LPU tarz\u0131 y\u00fcr\u00fctme, a\u00e7\u0131k bellek yerle\u015fimi gerektirirken CUDA \u00e7ekirdekleri donan\u0131m soyutlamas\u0131 \u00fczerine in\u015fa ediliyor. Bu iki yakla\u015f\u0131m\u0131n uyumlu h\u00e2le getirilmesi, Nvidia a\u00e7\u0131s\u0131ndan yaln\u0131zca donan\u0131m de\u011fil, yaz\u0131l\u0131m taraf\u0131nda da optimizasyon gerektirecek. Ancak\u00a0LPU entegrasyonu ger\u00e7ekle\u015firse, yaln\u0131zca ham performans de\u011fil, yapay zek\u00e2 \u00e7\u0131kar\u0131m i\u015f y\u00fcklerinde mimari d\u00fczeyde yeni bir sayfa a\u00e7abilir.<\/p>\n\n

<\/span><\/p>\n

Kaynak :\u00a0https:\/\/www.donanimhaber.com\/nvidia-feynman-mimarisi-geliyor-lpu-x3d-ve-dahasi–200242<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Nvidia’n\u0131n gelecek nesil Feynman GPU mimarisiyle birlikte \u00e7\u0131kar\u0131m odakl\u0131 \u00f6nemli bir de\u011fi\u015fime gidebilece\u011fi konu\u015fuluyor. Aktar\u0131lan bilgilere g\u00f6re \u015firket, 3D yap\u0131ya ek olarak Groq’a ait LPU (Language Processing Unit) birimlerini mimariye entegre …<\/p>\n","protected":false},"author":1,"featured_media":115716,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[523],"class_list":["post-115715","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-gpu"],"_links":{"self":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/115715","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/comments?post=115715"}],"version-history":[{"count":1,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/115715\/revisions"}],"predecessor-version":[{"id":115719,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/115715\/revisions\/115719"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media\/115716"}],"wp:attachment":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media?parent=115715"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/categories?post=115715"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/tags?post=115715"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}