{"id":2874,"date":"2025-04-14T17:21:13","date_gmt":"2025-04-14T09:21:13","guid":{"rendered":"https:\/\/aws-oncloudai.com\/?p=2874"},"modified":"2025-04-14T17:21:13","modified_gmt":"2025-04-14T09:21:13","slug":"ybgxjbjslafnatl","status":"publish","type":"post","link":"https:\/\/aws-oncloudai.com\/zh_hk\/ybgxjbjslafnatl\/","title":{"rendered":"\u64c1\u62b1\u9ad8\u6027\u50f9\u6bd4\u904b\u7b97\u529b\uff1aAmazon Graviton\u8ce6\u80fdAI\u63a8\u7406"},"content":{"rendered":"<p>\u96a8\u8457\u751f\u6210\u5f0fAI\u7684\u767c\u5c55\uff0c\u5927\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09\u63a8\u7406\u6210\u70ba\u4f01\u696d\u5728\u5efa\u69cb\u667a\u6167\u61c9\u7528\u6642\u7684\u6838\u5fc3\u7b97\u529b\u9700\u6c42\u4e4b\u4e00\u3002\u76f8\u8f03\u65bc\u8a13\u7df4\u74b0\u7bc0\uff0c\u63a8\u7406\u968e\u6bb5\u7684\u8ca0\u8377\u66f4\u5ee3\u6cdb\u3001\u5373\u6642\u6027\u8981\u6c42\u66f4\u9ad8\uff0c\u540c\u6642\u4e5f\u66f4\u91cd\u8996<strong>\u6027\u80fd\u8207\u6210\u672c\u7684\u5e73\u8861<\/strong>\u3002\u5728\u9019\u4e00\u80cc\u666f\u4e0b\uff0cAmazon Web Services\uff08AWS\uff09\u81ea\u7814\u7684 <strong>Graviton \u7cfb\u5217\u8655\u7406\u5668<\/strong>\uff0c\u6191\u85c9\u5176\u5353\u8d8a\u7684\u6027\u50f9\u6bd4\u8207\u5ee3\u6cdb\u7684\u517c\u5bb9\u6027\uff0c\u6b63\u6210\u70baAI\u63a8\u7406\u7684\u65b0\u9078\u64c7\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u4ec0\u9ebc\u662fAmazon Graviton\uff1f<\/h4>\n<p>Amazon Graviton \u662fAWS \u57fa\u65bcArm \u67b6\u69cb\u81ea\u4e3b\u7814\u767c\u7684\u8655\u7406\u5668\uff0c\u81f3\u4eca\u5df2\u8fed\u4ee3\u81f3 <strong>Graviton3<\/strong>\u3002\u76f8\u8f03\u65bc\u50b3\u7d71x86\u67b6\u69cb\u8655\u7406\u5668\uff0cGraviton \u5728\u4ee5\u4e0b\u5e7e\u500b\u65b9\u9762\u5177\u5099\u986f\u8457\u512a\u52e2\uff1a<\/p>\n<ul>\n<li><strong>\u9ad8\u6027\u50f9\u6bd4<\/strong>\uff1a\u76f8\u8f03\u65bc\u540c\u985e\u5be6\u4f8b\uff0c\u6027\u50f9\u6bd4\u63d0\u5347\u53ef\u9054<strong>40%<\/strong>\uff1b<\/li>\n<li><strong>\u4f4e\u529f\u8017<\/strong>\uff1a\u66f4\u5c11\u7684\u80fd\u8017\u5e36\u4f86\u66f4\u7da0\u8272\u7684\u8a08\u7b97\uff1b<\/li>\n<li><strong>\u591a\u6a23\u5316\u5be6\u4f8b\u985e\u578b<\/strong>\uff1a\u652f\u63f4\u5f9e\u901a\u7528\u578b\u3001\u904b\u7b97\u578b\u5230\u8a18\u61b6\u9ad4\u6700\u4f73\u5316\u578b\u7b49\u591a\u7a2e\u5be6\u4f8b\uff0c\u9748\u6d3b\u9069\u914dAI\u5834\u666f\uff1b<\/li>\n<li><strong>AWS \u539f\u751f\u6700\u4f73\u5316<\/strong>\uff1a\u8207Amazon EC2\u3001ECS\u3001EKS\u7b49\u670d\u52d9\u6df1\u5ea6\u96c6\u6210\uff0c\u958b\u7bb1\u5373\u7528\u3002<\/li>\n<\/ul>\n<p>&nbsp;<\/p>\n<h4>Graviton + AI\u63a8\u7406\uff1a\u70ba\u4ec0\u9ebc\u662f\u7d55\u914d\uff1f<\/h4>\n<p>\u96d6\u7136GPU\u5728AI\u8a13\u7df4\u9818\u57df\u4ecd\u5360\u4e3b\u5c0e\u5730\u4f4d\uff0c\u4f46\u5728<strong>\u63a8\u7406\u968e\u6bb5<\/strong>\uff0c\u5c24\u5176\u662f\u908a\u7de3\u90e8\u7f72\u3001\u5c0f\u6a21\u578b\u843d\u5730\u3001\u6279\u6b21\u670d\u52d9\u547c\u53eb\u7b49\u5834\u666f\uff0cCPU\u66f4\u5177\u5f48\u6027\u8207\u6210\u672c\u512a\u52e2\u3002<\/p>\n<h5>1.\u4f4e\u6210\u672c\u5927\u898f\u6a21\u90e8\u7f72<\/h5>\n<p>\u5728\u5927\u6a21\u578b\u5fae\u8abf\u6216\u91cf\u5316\u58d3\u7e2e\u5f8c\uff0c\u8a31\u591a\u5834\u666f\u4e0d\u518d\u4f9d\u8cf4GPU\u5f37\u7b97\u529b\uff0c\u800c\u662f\u8ffd\u6c42<strong>\u4e26\u767c\u6548\u80fd\u8207\u55ae\u6b21\u547c\u53eb\u6210\u672c<\/strong>\u3002 Graviton\u5177\u5099\u66f4\u9ad8\u7684\u6bcf\u7f8e\u5143\u6548\u80fd\u6bd4\uff0c\u80fd\u5920\u4ee5\u66f4\u4f4e\u7684\u6210\u672c\u5b8c\u6210\u63a8\u7406\u4efb\u52d9\uff0c\u9069\u5408\u5efa\u69cb\u5927\u898f\u6a21LLM\u670d\u52d9\u53e2\u96c6\u3002<\/p>\n<h5>2.\u652f\u6301\u4e3b\u6d41AI\u6846\u67b6<\/h5>\n<p>\u76ee\u524d\uff0cAWS\u5b98\u65b9\u63d0\u4f9b\u4e86\u91dd\u5c0dGraviton \u7684\u63a8\u7406\u52a0\u901f\u512a\u5316\uff0c\u5305\u62ec\uff1a<\/p>\n<ul>\n<li><strong>TensorFlow\u3001PyTorch<\/strong>\uff1a\u652f\u63f4Neon\u6307\u4ee4\u96c6\u3001XLA\u7de8\u8b6f\u5668\u7b49\u52a0\u901f\u65b9\u5f0f\uff1b<\/li>\n<li><strong>ONNX Runtime<\/strong>\uff1a\u5df2\u539f\u751f\u9069\u914dGraviton\uff0c\u5728\u591a\u57f7\u884c\u7dd2\u57f7\u884c\u4e0a\u8868\u73fe\u512a\u7570\uff1b<\/li>\n<li><strong>Hugging Face Transformers<\/strong>\uff1a\u53ef\u900f\u904eOptimum Neuron\u7b49\u5de5\u5177\u642d\u914d\u4f7f\u7528\uff1b<\/li>\n<li><strong>Amazon SageMaker<\/strong>\uff1a\u63d0\u4f9bGraviton\u5be6\u4f8b\u652f\u6301\uff0c\u7121\u9700\u5e95\u5c64\u914d\u7f6e\u5373\u53ef\u5b8c\u6210\u6a21\u578b\u90e8\u7f72\u3002<\/li>\n<\/ul>\n<h5>3.\u5be6\u6e2c\u7d50\u679c\u4ee4\u4eba\u9a5a\u8277<\/h5>\n<p>\u6839\u64daAWS\u5b98\u65b9\u6578\u64da\uff0c\u5728\u4f7f\u7528Amazon EC2 C7g\uff08\u57fa\u65bcGraviton3\uff09\u5be6\u4f8b\u9032\u884cBERT-base\u63a8\u7406\u6642\uff0c\u76f8\u8f03\u65bcC6i\uff08Intel\u67b6\u69cb\uff09\u5be6\u4f8b\uff0c\u5ef6\u9072\u964d\u4f4e <strong>20%<\/strong>\uff0c\u6bcf\u6b21\u63a8\u7406\u6210\u672c\u6e1b\u5c11 <strong>35%<\/strong>\uff0c\u5145\u5206\u8aaa\u660eGraviton\u5728AI\u63a8\u7406\u5834\u666f\u4e0b\u7684\u5be6\u969b\u512a\u52e2\u3002<\/p>\n<p>&nbsp;<\/p>\n<h4>\u4f01\u696d\u8a72\u5982\u4f55\u4e0a\u624b\uff1f<\/h4>\n<p>\u60f3\u8981\u5c07AI\u63a8\u7406\u9077\u79fb\u81f3Graviton\u5be6\u4f8b\uff0c\u5176\u5be6\u4e26\u4e0d\u8907\u96dc\u3002\u4ee5\u4e0b\u662f\u901a\u7528\u6d41\u7a0b\uff1a<\/p>\n<ol>\n<li><strong>\u8a55\u4f30\u73fe\u6709\u6a21\u578b\u76f8\u5bb9\u6027<\/strong>\uff1a\u78ba\u8a8d\u662f\u5426\u652f\u63f4\u5728Arm\u67b6\u69cb\u4e0a\u57f7\u884c\uff0c\u6216\u662f\u5426\u6709ONNX\u6a21\u578b\u53ef\u8f49\uff1b<\/li>\n<li><strong>\u9078\u64c7\u5408\u9069\u5be6\u4f8b\u985e\u578b<\/strong>\uff1a\u5982\u8a08\u7b97\u5bc6\u96c6\u578b\uff08C7g\uff09\u3001\u8a18\u61b6\u9ad4\u6700\u4f73\u5316\uff08R7g\uff09\u3001\u901a\u7528\u578b\uff08M7g\uff09\u7b49\uff1b<\/li>\n<li><strong>\u4ee3\u78bc\u9069\u914d\u8207\u8abf\u512a<\/strong>\uff1a\u4f7f\u7528AWS\u63d0\u4f9b\u7684Graviton\u512a\u5316\u5de5\u5177\u5305\uff08\u5982Amazon CodeCatalyst\u3001Graviton Ready SDK\uff09\uff1b<\/li>\n<li><strong>\u90e8\u7f72\u81f3SageMaker\u6216ECS\/EKS\u7b49\u5e73\u53f0<\/strong>\uff1a\u5feb\u901f\u5efa\u69cb\u7aef\u5230\u7aef\u63a8\u7406\u670d\u52d9\uff1b<\/li>\n<li><strong>\u76e3\u63a7\u6548\u80fd\u8868\u73fe\u4e26\u6301\u7e8c\u512a\u5316<\/strong>\uff1a\u7d50\u5408CloudWatch\u3001Amazon CloudTrail\u7b49\u5de5\u5177\u9032\u884c\u6548\u80fd\u5206\u6790\u3002<\/li>\n<\/ol>\n<p>&nbsp;<\/p>\n<h4>\u603b\u7ed3<\/h4>\n<p>\u5728AI\u4e0d\u65b7\u767c\u5c55\u7684\u4eca\u5929\uff0c\u4f01\u696d\u4e0d\u61c9\u518d\u5c40\u9650\u65bc\u50b3\u7d71\u786c\u9ad4\u8def\u5f91\u3002 Amazon Graviton \u4f5c\u70ba\u4e00\u6b3e\u9ad8\u6027\u50f9\u6bd4\u3001\u7da0\u8272\u7bc0\u80fd\u3001\u76f8\u5bb9\u5ee3\u6cdb\u7684CPU\u8655\u7406\u5668\uff0c\u6b63\u9010\u6f38\u6210\u70ba <strong>AI\u63a8\u7406\u65b0\u57fa\u77f3<\/strong>\u3002\u4e0d\u8ad6\u662f\u5efa\u7acb\u751f\u6210\u5f0fAI\u61c9\u7528\uff0c\u9084\u662f\u5728\u73fe\u6709\u6a21\u578b\u57fa\u790e\u4e0a\u8ffd\u6c42\u66f4\u4f4e\u7684\u90e8\u7f72\u6210\u672c\uff0cGraviton\u90fd\u503c\u5f97\u6210\u70ba\u4f60\u7684\u65b0\u9078\u9805\u3002<\/p>","protected":false},"excerpt":{"rendered":"<p>\u96a8\u8457\u751f\u6210\u5f0fAI\u7684\u767c\u5c55\uff0c\u5927\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09\u63a8\u7406\u6210\u70ba\u4f01\u696d\u5728\u5efa\u69cb\u667a\u6167\u61c9\u7528\u6642\u7684\u6838\u5fc3\u7b97\u529b\u9700\u6c42\u4e4b\u4e00\u3002\u76f8\u8f03\u65bc\u8a13\u7df4\u74b0\u7bc0\uff0c\u63a8\u7406\u968e\u6bb5\u7684\u8ca0\u8377\u66f4\u5ee3\u6cdb\u3001\u5373\u6642\u6027\u8981\u6c42\u66f4\u9ad8\uff0c\u540c\u6642\u4e5f\u66f4\u91cd\u8996\u6548\u80fd\u8207\u6210\u672c\u7684\u5e73\u8861\u3002\u5728\u9019\u4e00\u80cc\u666f\u4e0b\uff0cAmazon Web Services\uff08AWS\uff09\u81ea\u7814\u7684Graviton \u7cfb\u5217\u8655\u7406\u5668\uff0c\u6191\u85c9\u5176\u5353\u8d8a\u7684\u6027\u50f9\u6bd4\u8207\u5ee3\u6cdb\u7684\u517c\u5bb9\u6027\uff0c\u6b63\u6210\u70baAI\u63a8\u7406\u7684\u65b0\u9078\u64c7\u3002<\/p>","protected":false},"author":1,"featured_media":2877,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[65],"tags":[],"class_list":["post-2874","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technical-sharing"],"_links":{"self":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/2874","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/comments?post=2874"}],"version-history":[{"count":0,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/posts\/2874\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media\/2877"}],"wp:attachment":[{"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/media?parent=2874"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/categories?post=2874"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aws-oncloudai.com\/zh_hk\/wp-json\/wp\/v2\/tags?post=2874"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}