{"id":24452,"date":"2024-12-05T02:56:20","date_gmt":"2024-12-04T18:56:20","guid":{"rendered":"https:\/\/www.1ai.net\/?p=24452"},"modified":"2024-12-04T21:56:48","modified_gmt":"2024-12-04T13:56:48","slug":"%e4%ba%9a%e9%a9%ac%e9%80%8a-aws-ai-%e8%ae%ad%e7%bb%83%e8%8a%af%e7%89%87-trainium2-%e5%ae%9e%e4%be%8b%e5%85%a8%e9%9d%a2%e5%8f%af%e7%94%a8%ef%bc%8c%e5%85%ac%e5%b8%83%e4%b8%8b%e4%bb%a3-3nm-trainium3","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/24452.html","title":{"rendered":"Amazon AWS AI Training Chip Trainium2 Instances Fully Available, Announces Next-Gen 3nm Trainium3"},"content":{"rendered":"

\u4e9a\u9a6c\u900a<\/a> AWS \u4eca\u65e5\u5ba3\u5e03\uff0c\u57fa\u4e8e\u5176\u5185\u90e8\u56e2\u961f\u6240\u5f00\u53d1 AI \u8bad\u7ec3\u82af\u7247<\/a> Trainium2 \u7684 Trn2 \u5b9e\u4f8b\u5e7f\u6cdb\u53ef\u7528\uff0c\u5e76\u63a8\u51fa\u4e86 Trn2 UltraServer \u5927\u578b AI \u8bad\u7ec3\u7cfb\u7edf\uff0c\u540c\u65f6\u8fd8\u53d1\u5e03\u4e86\u4e0b\u4ee3\u66f4\u5148\u8fdb\u7684 3nm \u5236\u7a0b Trainium3 \u82af\u7247\u3002<\/p>\n

\"6bb9ab36j00snz2pj005od000uj00knp\"<\/p>\n

\u5355\u4e2a Trn2 \u5b9e\u4f8b\u5305\u542b 16 \u9897 Trainium2 \u82af\u7247\uff0c\u5404\u82af\u7247\u95f4\u91c7\u7528\u8d85\u9ad8\u901f\u9ad8\u5e26\u5bbd\u4f4e\u5ef6\u8fdf\u00a0NeuronLink \u4e92\u8054\uff0c\u53ef\u63d0\u4f9b 20.8\u00a0petaflops \u7684\u5cf0\u503c\u7b97\u529b\uff0c\u9002\u5408\u6570 B \u53c2\u6570\u5927\u5c0f\u6a21\u578b\u7684\u8bad\u7ec3\u548c\u90e8\u7f72\u3002<\/p>\n

\u4e9a\u9a6c\u900a\u5ba3\u79f0 Trn2 \u5b9e\u4f8b\u4e0e\u5f53\u524d\u4e00\u4ee3\u57fa\u4e8e GPU \u7684 EC2 P5e \u548c P5en \u5b9e\u4f8b\u76f8\u6bd4\u6027\u4ef7\u6bd4\u63d0\u9ad8\u4e86 30-40%<\/strong>\u3002<\/p>\n

\u800c\u66f4\u5927\u89c4\u6a21\u7684 Trn2 UltraServer \u5219\u7528 NeuronLink \u805a\u5408\u4e86 4 \u53f0 Trn2 \u670d\u52a1\u5668\uff0c\u603b\u5171\u5305\u542b 64 \u9897 Trainium2 \u82af\u7247\uff0c\u7b97\u529b\u5cf0\u503c\u8fdb\u4e00\u6b65\u7ebf\u6027\u6269\u5c55\u5230 83.2\u00a0petaflops\uff0c\u80fd\u6ee1\u8db3\u76ee\u524d\u4e16\u754c\u4e0a\u6700\u5927\u89c4\u6a21\u6a21\u578b\u7684\u8bad\u7ec3\u548c\u90e8\u7f72\u9700\u6c42<\/strong>\u3002<\/p>\n

\u4e9a\u9a6c\u900a\u8fd8\u6b63\u4e0e\u5176\u6295\u8d44\u7684 AI \u6a21\u578b\u4f01\u4e1a Anthropic \u5408\u4f5c\uff0c\u5171\u540c\u6784\u5efa\u4e00\u4e2a\u540d\u4e3a Project Rainier \u7684 EC2\u00a0UltraCluster \u5de8\u578b\u8ba1\u7b97\u96c6\u7fa4\uff0c\u8be5\u96c6\u7fa4\u5305\u542b\u5927\u91cf Trn2 UltraServer\uff0c\u603b\u5171\u62e5\u6709\u6570\u5341\u4e07\u9897\u00a0Trainium2 \u82af\u7247<\/strong>\u3002<\/p>\n

1AI\u83b7\u6089\uff0c\u8be5\u96c6\u7fa4\u5b8c\u6210\u540e\u6709\u671b\u6210\u4e3a\u8fc4\u4eca\u4e3a\u6b62\u516c\u5f00\u7684\u6700\u5927 AI \u8ba1\u7b97\u96c6\u7fa4<\/strong>\uff0c\u6574\u4f53\u7b97\u529b\u8fbe Anthropic \u76ee\u524d\u7528\u4e8e\u8bad\u7ec3\u6700\u5148\u8fdb Claude \u6a21\u578b\u6240\u9700\u91cf\u7684 5 \u500d\u4ee5\u4e0a\u3002<\/p>\n

\u4e9a\u9a6c\u900a AWS \u8fd8\u516c\u5e03\u4e86\u4e0b\u4ee3 Trainium3 AI \u8bad\u7ec3\u82af\u7247\uff0c\u8fd9\u4e5f\u662f AWS \u9996\u6b3e\u91c7\u7528 3nm \u5236\u7a0b\u7684\u82af\u7247\u4ea7\u54c1\u3002\u4e9a\u9a6c\u900a\u8868\u793a\u57fa\u4e8e Trainium3 \u7684 UltraServer \u6027\u80fd\u53ef\u8fbe Trn2 UltraServer \u7684 4 \u500d<\/strong>\uff0c\u800c\u9996\u6279\u57fa\u4e8e Trainium3 \u7684\u5b9e\u4f8b\u9884\u8ba1\u5c06\u4e8e 2025 \u5e74\u5e95\u63a8\u51fa\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

Amazon AWS today announced the widespread availability of Trn2 instances based on its in-house team-developed AI training chip, Trainium2, with the launch of the Trn2 UltraServer large-scale AI training system, as well as the release of the next generation of the even more advanced 3nm-processed Trainium3 chip. A single Trn2 instance consists of 16 Trainium2 chips, each interconnected by an ultra-fast, high-bandwidth, low-latency NeuronLink interconnect to deliver 20.8 petaflops of peak power, suitable for training and deploying several B-parameter-sized models. Amazon claims that Trn2 instances are comparable to the current generation of GPU-based EC2<\/p>","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[146],"tags":[370,238],"collection":[],"class_list":["post-24452","post","type-post","status-publish","format-standard","hentry","category-news","tag-370","tag-238"],"acf":[],"_links":{"self":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/24452","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/comments?post=24452"}],"version-history":[{"count":0,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/posts\/24452\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/media?parent=24452"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/categories?post=24452"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/tags?post=24452"},{"taxonomy":"collection","embeddable":true,"href":"https:\/\/www.1ai.net\/en\/wp-json\/wp\/v2\/collection?post=24452"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}