{"id":11538,"date":"2022-11-10T15:54:40","date_gmt":"2022-11-10T07:54:40","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=11538"},"modified":"2022-11-10T15:54:42","modified_gmt":"2022-11-10T07:54:42","slug":"nvidia-hopper-%e8%88%87-ampere-gpu-%e5%9c%a8%e4%ba%ba%e5%b7%a5%e6%99%ba%e6%85%a7%e8%a8%93%e7%b7%b4%e5%9f%ba%e6%ba%96%e6%b8%ac%e8%a9%a6%e4%b8%ad%e6%a9%ab%e6%8e%83%e5%8d%83%e8%bb%8d","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=11538","title":{"rendered":"NVIDIA Hopper\u8207Ampere GPU\u5728\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u57fa\u6e96\u6e2c\u8a66\u4e2d\u6a6b\u6383\u5343\u8ecd"},"content":{"rendered":"\n

\u5728\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u7684\u7522\u696d\u6a19\u6e96\u6e2c\u8a66\u4e2d\uff0cNVIDIA H100 Tensor\u00a0\u6838\u5fc3\u00a0GPU\uff0c\u5728\u8655\u7406\u4f01\u696d\u4f5c\u696d\u8ca0\u8f09\u5275\u4e0b\u4e16\u754c\u7d00\u9304\uff0c\u800c\u00a0A100\u00a0\u5247\u63d0\u9ad8\u4e86\u9ad8\u6548\u80fd\u904b\u7b97\u7684\u6a19\u6e96<\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

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

NVIDIA H100 Tensor\u00a0\u6838\u5fc3\u00a0GPU<\/a>\u00a0\u5728\u521d\u767b\u5834<\/a>\u5237\u65b0\u00a0MLPerf\u00a0\u63a8\u8ad6\u57fa\u6e96\u6e2c\u8a66\u7d00\u9304\u7684\u5169\u500b\u6708\u5f8c\uff0c\u53c8\u5728\u7522\u696d\u7d44\u7e54\u7684\u4eba\u5de5\u667a\u6167\u00a0(AI)\u00a0\u8a13\u7df4\u6700\u65b0\u6e2c\u8a66\u4e2d\uff0c\u5275\u4e0b\u8655\u7406\u4f01\u696d\u00a0AI\u00a0\u4f5c\u696d\u8ca0\u8f09\u7684\u5168\u65b0\u4e16\u754c\u7d00\u9304\u3002\u9019\u4e9b\u6e2c\u8a66\u7d50\u679c\u986f\u793a\uff0c\u5c0d\u65bc\u5efa\u7acb\u8207\u90e8\u7f72\u6700\u5148\u9032\u00a0AI\u00a0\u6a21\u578b\u4ee5\u8981\u6c42\u6700\u9ad8\u8868\u73fe\u7684\u7528\u6236\u4f86\u8aaa\uff0cH100\u00a0\u662f\u6700\u4f73\u9078\u64c7\u3002<\/p>\n\n\n\n

MLPerf \u662f\u7522\u696d\u7528\u65bc\u8861\u91cf AI \u6548\u80fd\u7684\u6a19\u6e96\u3002\u5305\u62ec\u4e9e\u99ac\u905c (Amazon)\u3001Arm\u3001\u767e\u5ea6 (Baidu)\u3001Google\u3001\u54c8\u4f5b\u5927\u5b78 (Harvard University)\u3001\u82f1\u7279\u723e (Intel)\u3001Meta\u3001\u5fae\u8edf (Microsoft)\u3001\u53f2\u4e39\u4f5b\u5927\u5b78 (Stanford University) \u548c\u591a\u502b\u591a\u5927\u5b78 (the University of Toronto) \u5728\u5167\u7684\u773e\u591a\u696d\u8005\u8207\u5b78\u8853\u55ae\u4f4d\u7686\u5c0d\u9019\u9805\u6a19\u6e96\u8868\u9054\u652f\u6301\u3002<\/p>\n\n\n\n

\u5728\u4eca\u5929\u767c\u5e03\u7684 MLPerf \u76f8\u95dc\u57fa\u6e96\u6e2c\u8a66\u4e2d\uff0cNVIDIA A100 Tensor \u6838\u5fc3 GPU<\/a> \u63d0\u9ad8\u4e86\u53bb\u5e74<\/a>\u5728\u9ad8\u6548\u80fd\u904b\u7b97 (HPC) \u7acb\u4e0b\u7684\u6a19\u6e96\u3002<\/p>\n\n\n\n

\"\"
\u25b2NVIDIA H100 GPU\u00a0\u9996\u6b21\u63d0\u4ea4\u00a0MLPerf\u00a0\u8a13\u7df4\u6210\u7e3e\u6642\uff0c\u901f\u5ea6\u6bd4\u00a0A100 GPU\u00a0\u5feb\u00a06.7\u00a0\u500d<\/figcaption><\/figure>\n\n\n\n

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

H100 GPU (\u63a1\u7528 Hopper \u67b6\u69cb) \u63d0\u9ad8\u4e86 MLPerf \u8a13\u7df4\u4e2d\u55ae\u4e00\u52a0\u901f\u5668\u7684\u6548\u80fd\u6a19\u6e96\u3002\u5728\u9996\u6b21\u5c07 H100 GPU \u7528\u65bc MLPerf \u8a13\u7df4\u6642\uff0c\u8868\u73fe\u6bd4\u524d\u4e00\u4ee3 GPU \u9ad8\u51fa 6.7 \u500d\uff1b\u800c\u4ee5\u540c\u6a23\u7684\u6bd4\u8f03\u57fa\u790e\u4f86\u770b\uff0c\u7531\u65bc\u8edf\u9ad4\u7684\u9032\u6b65\uff0c\u76ee\u524d A100 GPU \u7684\u8655\u7406\u80fd\u529b\u53c8\u9ad8\u51fa 2.5 \u500d\u3002<\/p>\n\n\n\n

\u6709\u90e8\u5206\u539f\u56e0\u51fa\u81ea\u65bc\u5b83\u7684 Transformer \u5f15\u64ce<\/a>\uff0cHopper \u67b6\u69cb\u5728\u8a13\u7df4\u7528\u65bc\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u7684\u71b1\u9580 BERT \u6a21\u578b\u6642\u6709\u6975\u70ba\u51fa\u8272\u7684\u8868\u73fe\u3002BERT \u662f MLPerf AI \u6a21\u578b\u4e2d\u898f\u6a21\u6700\u5927\u3001\u5c0d\u8655\u7406\u6548\u80fd\u8981\u6c42\u6700\u56b4\u82db\u7684\u6a21\u578b\u4e4b\u4e00\u3002<\/p>\n\n\n\n

\u7531\u65bc\u9019\u4e9b\u57fa\u6e96\u6db5\u84cb\u7576\u4eca\u6700\u53d7\u6b61\u8fce\u7684\u00a0AI\u00a0\u4f5c\u696d\u8ca0\u8f09\uff0c\u50cf\u662f\u96fb\u8166\u8996\u89ba\u3001\u81ea\u7136\u8a9e\u8a00\u8655\u7406\u3001\u63a8\u85a6\u7cfb\u7d71\u3001\u5f37\u5316\u5b78\u7fd2\u7b49\uff0cMLPerf\u00a0\u7684\u6e2c\u8a66\u7d50\u679c\u8b93\u7528\u6236\u6709\u4fe1\u5fc3\u505a\u51fa\u660e\u667a\u7684\u8cfc\u8cb7\u6c7a\u5b9a\u3002\u9019\u4e9b\u6e2c\u8a66\u7d93\u904e\u696d\u754c\u5be9\u6838\uff0c\u503c\u5f97\u7528\u6236\u4fe1\u4efb\u3002<\/p>\n\n\n\n

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

A100 GPU <\/strong>\u5728\u9ad8\u6548\u80fd\u904b\u7b97\u9054\u5230\u65b0\u9ad8\u5cf0<\/strong><\/strong><\/h3>\n\n\n\n

\u5728\u7368\u7acb\u7684 MLPerf HPC \u57fa\u6e96\u6e2c\u8a66\uff0c\u5728\u8d85\u7d1a\u96fb\u8166\u4e0a\u904b\u884c\u56b4\u82db\u7684\u79d1\u5b78\u4f5c\u696d\u8ca0\u8f09\u6642\uff0cA100 GPU \u5728 AI \u8a13\u7df4\u6a21\u578b\u6240\u6709\u6e2c\u8a66\u9805\u76ee\u4e2d\u7684\u8868\u73fe\u6a6b\u6383\u5343\u8ecd\u3002\u9019\u4e9b\u7d50\u679c\u986f\u793a NVIDIA AI \u5e73\u53f0\u6709\u80fd\u529b\u653b\u514b\u4e16\u754c\u4e0a\u6700\u8271\u96e3\u7684\u6280\u8853\u6311\u6230\u3002<\/p>\n\n\n\n

\u50cf\u662f A100 GPU \u5728 CosmoFlow \u6e2c\u8a66\u4e2d\u8a13\u7df4 AI \u6a21\u578b\uff0c\u6bd4\u5169\u5e74\u524d\u7b2c\u4e00\u8f2a MLPerf HPC \u6e2c\u8a66\u7684\u6700\u4f73\u7d50\u679c\u9084\u8981\u5feb 9 \u500d\u3002\u5728\u540c\u6a23\u7684\u4f5c\u696d\u8ca0\u8f09\u4e2d\uff0cA100 \u7684\u6bcf\u584a\u6676\u7247\u8655\u7406\u91cf\u4ea6\u9a5a\u4eba\u5730\u6bd4\u5176\u4ed6\u7522\u54c1\u9ad8\u51fa 66 \u500d\u3002<\/p>\n\n\n\n

HPC\u00a0\u57fa\u6e96\u662f\u7528\u65bc\u5929\u9ad4\u7269\u7406\u5b78\u3001\u5929\u6c23\u9810\u6e2c\u53ca\u5206\u5b50\u52d5\u529b\u5b78\u7684\u5de5\u4f5c\u8a13\u7df4\u6a21\u578b\uff0c\u85e5\u7269\u958b\u767c\u7b49\u8a31\u591a\u6280\u8853\u9818\u57df\u540c\u6a23\u63a1\u7528\u00a0AI\u00a0\u4f86\u63a8\u52d5\u79d1\u5b78\u767c\u5c55\u3002<\/p>\n\n\n\n

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

\"\"
\u25b2\u5728\u5168\u7403\u7684\u6e2c\u8a66\u4e2d\uff0cA100 GPU\u00a0\u5728\u8a13\u7df4\u901f\u5ea6\u8207\u8655\u7406\u91cf\u90fd\u8655\u65bc\u9818\u5148\u5730\u4f4d<\/figcaption><\/figure>\n\n\n\n

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

\u4e9e\u6d32\u3001\u6b50\u6d32\u53ca\u7f8e\u570b\u7684\u8d85\u7d1a\u96fb\u8166\u4e2d\u5fc3\u53c3\u52a0\u6700\u65b0\u4e00\u8f2a\u7684\u00a0MLPerf HPC\u00a0\u6e2c\u8a66\u3002\u5728\u9996\u6b21\u53c3\u52a0\u00a0DeepCAM\u00a0\u57fa\u6e96\u6e2c\u8a66\u6642\uff0c\u6234\u723e\u79d1\u6280\u00a0(Dell Technologies)\u00a0\u4f7f\u7528\u00a0NVIDIA A100 GPU\u00a0\u62ff\u4e0b\u4eae\u773c\u6210\u7e3e\u3002<\/p>\n\n\n\n

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

\u7121\u8207\u502b\u6bd4\u7684\u751f\u614b\u7cfb<\/strong><\/strong><\/h3>\n\n\n\n

\u5728\u4f01\u696d AI \u8a13\u7df4\u57fa\u6e96\u6e2c\u8a66\u65b9\u9762\uff0c\u5305\u62ecMicrosoft Azure \u96f2\u7aef\u670d\u52d9<\/a>\u5728\u5167\u7684 11 \u5bb6\u516c\u53f8\u4f7f\u7528 NVIDIA A100\u3001A30 \u548c A40 GPU \u63d0\u4ea4\u6e2c\u8a66\u7d50\u679c\u3002\u83ef\u78a9 (ASUS)\u3001\u6234\u723e\u79d1\u6280\u3001\u5bcc\u58eb\u901a (Fujitsu)\u3001\u6280\u5609 (GIGABYTE)\u3001\u6167\u8207\u79d1\u6280 (Hewlett Packard Enterprise)\u3001\u806f\u60f3 (Lenovo) \u53ca\u7f8e\u8d85\u5fae (Supermicro) \u7b49\u7cfb\u7d71\u88fd\u9020\u5546\uff0c\u5171\u4f7f\u7528\u4e5d\u5957 NVIDIA \u8a8d\u8b49\u7cfb\u7d71<\/a>\u63d0\u4ea4\u6e2c\u8a66\u7d50\u679c\u3002<\/p>\n\n\n\n

\u5728\u6700\u65b0\u4e00\u8f2a\u7684\u6e2c\u8a66\u4e2d\uff0c\u81f3\u5c11\u6709\u4e09\u5bb6\u516c\u53f8\u8207 NVIDIA \u4e00\u6a23\u63d0\u4ea4\u6240\u6709 MLPerf \u8a13\u7df4\u4f5c\u696d\u8ca0\u8f09\u7684\u516b\u9805\u6e2c\u8a66\u7d50\u679c\u3002\u771f\u5be6\u74b0\u5883\u4e2d\u7684\u61c9\u7528\u7a0b\u5f0f\u5f80\u5f80\u9700\u8981\u4e00\u5957\u6db5\u84cb\u591a\u7a2e AI \u6a21\u578b\u7684\u65b9\u6848\uff0c\u56e0\u6b64\u901a\u7528\u6027\u975e\u5e38\u91cd\u8981\u3002<\/p>\n\n\n\n

NVIDIA\u00a0\u7684\u5408\u4f5c\u5925\u4f34\u8a8d\u77e5\u5230\u5ba2\u6236\u6703\u4f7f\u7528\u00a0MLPerf\u00a0\u9019\u9805\u5bf6\u8cb4\u5de5\u5177\u4f86\u8a55\u4f30\u00a0AI\u00a0\u5e73\u53f0\u548c\u4f9b\u61c9\u5546\uff0c\u56e0\u6b64\u53c3\u8207\u00a0MLPerf\u00a0\u57fa\u6e96\u6e2c\u8a66\u3002<\/p>\n\n\n\n

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

\u80cc\u5f8c\u7684\u79d8\u5bc6<\/strong><\/h3>\n\n\n\n

NVIDIA AI \u5e73\u53f0\u63d0\u4f9b\u4e00\u500b\u5f9e\u6676\u7247\u5230\u7cfb\u7d71\u3001\u8edf\u9ad4\u8207\u670d\u52d9\u7684\u5b8c\u6574\u5806\u758a\uff0c\u53ef\u6301\u7e8c\u4e0d\u65b7\u5730\u63d0\u5347\u904b\u4f5c\u8868\u73fe\u3002\u4f8b\u5982\u5728\u6700\u65b0\u7684 HPC \u6e2c\u8a66\u4e2d\uff0c\u53c3\u8207\u8005\u63d0\u51fa\u7684\u7d50\u679c\u4f7f\u7528\u4e86\u4e00\u7bc7\u6280\u8853\u6587\u7ae0\u4e2d\u63cf\u8ff0\u7684\u4e00\u5957\u8edf\u9ad4\u6700\u4f73\u5316\u9805\u76ee\u548c\u6280\u8853\u3002\u9019\u4e9b\u56e0\u7d20\u52a0\u8d77\u4f86\u5c07\u4e00\u500b\u57fa\u6e96\u6e2c\u8a66\u7684\u57f7\u884c\u6642\u9593\u7e2e\u77ed 5 \u500d\uff0c\u5f9e\u539f\u672c\u7684 101 \u5206\u9418\u7e2e\u77ed\u81f3 22 \u5206\u9418\u3002<\/p>\n\n\n\n

\u7b2c\u4e8c\u7bc7\u6587\u7ae0<\/a>\u4ecb\u7d39 NVIDIA \u5982\u4f55\u91dd\u5c0d\u4f01\u696d AI \u57fa\u6e96\u6e2c\u8a66\u6700\u4f73\u5316\u5176\u5e73\u53f0\u3002\u4f8b\u5982\uff0c\u6211\u5011\u4f7f\u7528 NVIDIA DALI<\/a> \u70ba\u904b\u7b97\u8996\u89ba\u57fa\u6e96\u6709\u6548\u5730\u8f09\u5165\u8207\u9810\u8655\u7406\u8cc7\u6599\u3002<\/p>\n\n\n\n

\u7528\u65bc\u6e2c\u8a66\u7684\u6240\u6709\u8edf\u9ad4\u53ef\u5728 MLPerf \u8cc7\u6e90\u5eab\u4e2d\u53d6\u5f97\uff0c\u4efb\u4f55\u4eba\u90fd\u80fd\u53d6\u5f97\u9019\u4e9b\u4e16\u754c\u7d1a\u7684\u6e2c\u8a66\u7d50\u679c\u3002NVIDIA \u4e0d\u65b7\u5c07\u9019\u4e9b\u6700\u4f73\u5316\u9805\u76ee\u7d0d\u5165\u53ef\u5728 GPU \u61c9\u7528\u7a0b\u5f0f\u8edf\u9ad4\u4e2d\u5fc3 NGC<\/a> \u53d6\u5f97\u7684\u5bb9\u5668\u5167\u3002<\/p>\n\n\n\n

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

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

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

\ud83d\udfe6\u73fe\u5728\u5c31\u52a0\u5165 ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718<\/a> \u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!<\/strong>
\"\ud83d\udfe9\"\u6211\u5011\u6709LINE TODAY<\/a>\u983b\u9053\u4e86\uff0c\u5feb\u4f86\u8ffd\u8e2a\u6211\u5011\u5427!!–\u6700\u65b0\u79d1\u6280\u65b0\u805e \u76e1\u5728\u4f60\u624b<\/strong><\/h4>\n\n\n\n

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

<\/p>\n","protected":false},"excerpt":{"rendered":"

\u5728\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u7684\u7522\u696d\u6a19\u6e96\u6e2c\u8a66\u4e2d\uff0cNV<\/p>\n","protected":false},"author":3,"featured_media":11532,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[3032,1287,71,642],"class_list":["post-11538","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-ampere-gpu","tag-hopper","tag-nvidia","tag-642"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/11538"}],"collection":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=11538"}],"version-history":[{"count":1,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/11538\/revisions"}],"predecessor-version":[{"id":11541,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/11538\/revisions\/11541"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/11532"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=11538"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=11538"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=11538"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}