NVIDIA AI Summit Japan 2024ã«ã¦çºè¡¨ããå 容ã§ãã ç»å£æ¥ï¼ 2024å¹´11æ13æ¥ ã¤ãã³ããªã³ã¯ï¼Â https://www.nvidia.com/ja-jp/events/ai-summit/ ç»å£åç»ï¼Â https://www.nvidia.com/ja-jp/onâ¦
ã¯ããã« ããã«ã¡ã¯ãITåºç¤ãããã¯ã¼ã¯ã°ã«ã¼ãã®å®å±ã¨ç³ãã¾ãã 主ã«ç¤¾å ã®ãããã¯ã¼ã¯ãCDN (Content Delivery Network)é¢é£ã®æ¥åãæ å½ãã¦ãã¾ãã ä»ã¾ã§ã®ããã°ã§ã¯ãããã¯ã¼ã¯ã°ã«ã¼ãã§æ¡ç¨ãã¦ãããµã¼ãã¹ãããããã¯ã¼ã¯ç§»è¡ã«ã¤ãã¦ãç´¹ä»ãã¦ãã¾ãããã ä»å㯠DeNA ã®ãããã¯ã¼ã¯éç¨ç£è¦ã§ä½¿ç¨ãã¦ããã¼ã«ããç´¹ä»ãããã¾ãã AWS BYOIP ã使ã£ãèªç¤¾ IPv4 éç¨ææ³ DeNA ã§ã® GCP ãããã¯ã¼ã¯éç¨ DeNA æ¬ç¤¾ç§»è»¢ã§ãããã¯ã¼ã¯æ§ç¯ã»ç§»è¡ä½æ¥ãå®æ½ãã¾ãã DeNA ã®ãããã¯ã¼ã¯ã«ã¤ã㦠ãããã¯ã¼ã¯ã¯é害ãå質ãæªãã¨è¤æ°ã®ãµã¼ãã¹ãã¦ã¼ã¶ã«å½±é¿ãä¸ãã¾ãã ãã®ããã24/365ä½å¶ã§ãæ¥åãåæ»ã«é²ãããããããªãããã¯ã¼ã¯ãéç¨ç£è¦ãããã¨ããããã¯ã¼ã¯ã°ã«ã¼ãã®åºæ¬æ¹éã¨ãªãã¾ãã ãã ãããªã½ã¼ã¹ã¯æéãª
ããã«ã¡ã¯ããããã®ãã¬ãã¸ç·¨éé¨ã§ãã 7æã«éå¬ãããJANOG54 Meetingã§ãçæAIåããããªãã¯ã¯ã©ã¦ããµã¼ãã¹ãã¤ãã£ã¦ã¿ã話ãã«ã¤ãã¦ããããã¤ã³ã¿ã¼ããã é«å³¯ èª ãããäºä¸ å¬è¦ãããå¹³ç° å¤§ç¥ãããç»å£ãã¾ããããã®å 容ãã¬ãã¼ããã¾ãã çæAIåãåºç¤ã«ã¤ãã¦ãã¾ãã¯æç³»åãç¨ãã¦æ¦è¦ã説æãã¾ãã2011å¹´ã«ããããã¤ã³ã¿ã¼ãããã¯åæµ·éç³ç©ã«èªç¤¾éå¶ã®ç³ç©ãã¼ã¿ã»ã³ã¿ãéæãã¾ããã2016å¹´9æã«åãã¦GPUã®ã³ã³ãã¥ã¼ãã£ã³ã°ãªã½ã¼ã¹ãæä¾ããããããã®å°ç¨ãµã¼ã é«ç«åãã·ãªã¼ãºã®ãµã¼ãã¹ãéå§ãã2020å¹´7æã«ã¯ãããã®å°ç¨ãµã¼ãã®æ°ãããããã¯ã¼ã¯åºç¤ããããã®å°ç¨ãµã¼ã PHYãã¨ãããµã¼ãã¹ãæä¾éå§ãã¾ãããããã«ã2021å¹´ã«ã¯2016å¹´ã«æä¾éå§ãããããã®å°ç¨ãµã¼ãé«ç«åã®GPUãµã¼ããä»®æ³åãããããã®ã¯ã©ã¦ãã«è¼ããããããã¯
éä¿¡ãããã¯ã¼ã¯ã®å½¹å²ã¯âãã¼ã¿ãä¼ããâãã¨ã ãã ããæ¡éãã®è¨ç®è½åãå¿ è¦ã¨ããçæAI/æ©æ¢°å¦ç¿ã®å ´ã§ã¯ãã®å¸¸èãå¤ããããããã¯ã¼ã¯ã«ãè¨ç®ãªã½ã¼ã¹ã®ä¸é¨ã¨ãã¦åãæ§è½ã»æ©è½ãæ±ããããã 2022å¹´ãâChatGPTå ¬éâã®å¹´ãªãã2023å¹´ã¯ããã®Open AIã追ããããè ãã¡ã®è©±é¡ã§æã¡åãã®å¹´ã ã£ããåç±³ã®ããã°ããã¯ã¯ãã¨ãããå½å ã®åããæ´»çºã ã NTTãã½ãããã³ã¯ãå¯å£«éãçç ãã次ã ã¨ãçæAIã®åºç¤ã¨ãªã大è¦æ¨¡è¨èªã¢ãã«ï¼LLMï¼ã®éçºã表æããµã¤ãã¼ã¨ã¼ã¸ã§ã³ãã¯5æãNECã¯7æã«èªç¤¾éçºããLLMã®å ¬éãæä¾ãéå§ããããã®å¾ãæ±äº¬å¤§å¦æ¾å°¾ç 究室ï¼æ¾å°¾è±ææï¼ãPreferred NetworksãLLMãå ¬éã2024å¹´3æã«ã¯ãNTTããtsuzumiãã®ããã¥ã¼ãäºå®ãã¦ããã ãè¨ç®ã®ããã®ãããã¯ã¼ã¯ã 第ä¸æ¬¡AIãã¼ã ã®è¡æ¹ã«ã¯ã¾ã ä¸éæãªé¨
Alibaba Cloudã®ã¨ã³ã¸ãã¢å ¼ç 究è ã§ããEnnan Zhaiæ°ããLLMãã¬ã¼ãã³ã°åããã¼ã¿ã»ã³ã¿ã¼ã®GPUãã¤ã¼ãµãããã§ç¸äºæ¥ç¶ããæ¹æ³ã«é¢ããç 究è«æãçºè¡¨ãã¾ããããã®æè¡ã¯ç¹ã«ãNVIDIAã®GPUç¸äºæ¥ç¶ãããã³ã«ã§ããNVLinkã§ã¯ãªãã¤ã¼ãµããããæ¡ç¨ãã¦ããç¹ã§æ¬æ°ã¨ããã¦ãã¾ãã Alibaba HPN: A Data Center Network for Large Language Model Training (PDFãã¡ã¤ã«)https://ennanzhai.github.io/pub/sigcomm24-hpn.pdf Alibaba Cloud ditches Nvidia's interconnect in favor of Ethernet â tech giant uses own High Performance Network t
å¹³ç´ ã¯ãæ ¼å¥ã®ãé«é ãè³ãåãã礼ç³ãä¸ãã¾ãã ãã®åº¦ã¯ãæ¨æºä»æ§ã¨ãªãã¾ãAWG24ï¼å°ä½å¾0.5ï½ï½ï¼ã«å ãã¦ãAWG26ï¼å°ä½å¾0.4ï½ï½ï¼ç´°å¾ä»æ§ãã©ã¤ã³ããããã¦ããã¾ãã æ大ãã£ãã«é·75ï½ã§ãããã¨ãããå製åã·ãªã¼ãºããSMART75ãã¨åä»ãã¦è²©å£²ãã¦ããã¾ãã Cat.6A AWG24ã±ã¼ãã«å¤å¾ã太ãæ½å·¥ãé£ãããããå°ãã§ãã±ã¼ãã«ãç´°ããããã¨ã§æ½å·¥æ§ã®åä¸ãå³ãã¾ããã æ¯éã¨ãã覧ããã ãããæ¤è¨ããã ããã¨å¹¸ãã§ãã
å¸å ´ã§ã¯ãå ã¢ã¸ã¥ã¼ã«ã¨GPUã®æ¯çãè¨ç®ããããã®è¤æ°ã®ã¢ããã¼ããåå¨ããä¸è²«æ§ã®ãªãçµæãããããã¦ãã¾ãããããã®éãã®ä¸»ãªåå ã¯ããã¾ãã¾ãªãããã¯ã¼ã¯æ§é ã«å®è£ ãããå ã¢ã¸ã¥ã¼ã«ã®æ°ã®å¤åã«ç±æ¥ãã¦ãã¾ããå¿ è¦ãªå ã¢ã¸ã¥ã¼ã«ã®æ£ç¢ºãªæ°éã¯ãããã¤ãã®éè¦ãªè¦å ã«ä¸»ã«ä¾åãã¦ãã¾ãã ãããã¯ã¼ã¯ã«ã¼ãã¢ã㫠主ã«2ã¤ã®ãããã¯ã¼ã¯ã«ã¼ããå«ã¾ãã¦ãããConnectX-6ï¼200Gb/sã主ã«A100ã¨ä½¿ç¨ãããï¼ã¨ConnectX-7ï¼400Gb/sã主ã«H100ã¨ä½¿ç¨ãããï¼ã§ãã åæã«ã次ä¸ä»£ã®ConnectX-8 800Gb/sã2024å¹´ã«ãªãªã¼ã¹ãããäºå®ã§ãã ã¹ã¤ããã¢ã㫠主ã«2種é¡ã®ã¹ã¤ãããå«ã¾ãã¦ãããQM 9700ã¹ã¤ããï¼32ãã¼ãOSFP 2x400Gb/sï¼ãããã¾ããåè¨64ãã£ã³ãã«ã®400Gb/sã®è»¢éé度ã¨ãåè¨51.2Tb/sã®ã¹ã«ã¼
ãªãªã¼ã¹ãé害æ å ±ãªã©ã®ãµã¼ãã¹ã®ãç¥ãã
ææ°ã®äººæ°ã¨ã³ããªã¼ã®é ä¿¡
å¦çãå®è¡ä¸ã§ã
j次ã®ããã¯ãã¼ã¯
kåã®ããã¯ãã¼ã¯
lãã¨ã§èªã
eã³ã¡ã³ãä¸è¦§ãéã
oãã¼ã¸ãéã
{{#tags}}- {{label}}
{{/tags}}