{"id":71605,"date":"2024-06-13T09:57:08","date_gmt":"2024-06-13T01:57:08","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=71605"},"modified":"2024-06-13T09:57:09","modified_gmt":"2024-06-13T01:57:09","slug":"%e6%93%b4%e5%b1%95%e5%88%b0%e6%96%b0%e9%ab%98%e5%ba%a6%ef%bc%9anvidia-mlperf%e8%a8%93%e7%b7%b4%e7%b5%90%e6%9e%9c%e5%b1%95%e7%a4%ba%e4%ba%86%e5%89%8d%e6%89%80%e6%9c%aa%e6%9c%89%e7%9a%84%e6%95%88","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=71605","title":{"rendered":"\u64f4\u5c55\u5230\u65b0\u9ad8\u5ea6\uff1aNVIDIA MLPerf\u8a13\u7df4\u7d50\u679c\u5c55\u793a\u4e86\u524d\u6240\u672a\u6709\u7684\u6548\u80fd\u548c\u5f48\u6027"},"content":{"rendered":"\n

NVIDIA\u5728\u6700\u65b0\u7684MLPerf\u8a13\u7df4\u57fa\u6e96\u6e2c\u8a66\u4e2d\u9054\u5230\u4e86\u7121\u8207\u502b\u6bd4\u7684\u6548\u80fd\u548c\u8fd1100%\u7684\u64f4\u5145\u6548\u7387<\/h3>\n\n\n\n

\u5168\u7aef NVIDIA \u52a0\u901f\u904b\u7b97\u5e73\u53f0\u5728\u6700\u65b0\u7684 MLPerf Training v4.0 \u57fa\u6e96\u6e2c\u8a66\u4e2d\u518d\u6b21\u5c55\u73fe\u51fa\u5353\u8d8a\u7684\u6548\u80fd\u3002<\/p>\n\n\n\n

\u8207\u53bb\u5e74 NVIDIA \u63d0\u4ea4\u5275\u65b0\u7d00\u9304\u7684\u6578\u64da\u76f8\u6bd4\uff0cNVIDIA \u5728\u57fa\u65bc GPT-3 175B \u7684\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09\u57fa\u6e96\u6e2c\u8a66\u4e2d\u7684\u8868\u73fe\uff0c\u63d0\u5347\u4e86\u4e09\u500d\u4ee5\u4e0a\u3002 NVIDIA \u4f7f\u7528\u914d\u5099 11,616 \u500b NVIDIA H100 Tensor Core GPU<\/a> \u4e26\u8207 NVIDIA Quantum-2 InfiniBand<\/a> \u7db2\u8def\u9023\u63a5\u7684\u4eba\u5de5\u667a\u6167\uff08AI\uff09\u8d85\u7d1a\u96fb\u8166\uff0c\u900f\u904e\u66f4\u5927\u898f\u6a21\uff08\u6bd4\u4e00\u5e74\u524d\u63d0\u4ea4\u7684 3,584 \u500b H100 GPU \u589e\u52a0\u4e09\u500d\u591a\uff09\u548c\u5ee3\u6cdb\u7684\u5168\u7aef\u5de5\u7a0b\u5be6\u73fe\u4e86\u9019\u9805\u975e\u51e1\u58ef\u8209\u3002<\/p>\n\n\n\n

\u7531\u65bc NVIDIA AI \u5e73\u53f0\u7684\u53ef\u64f4\u5c55\u6027\uff0cEos \u73fe\u5728\u53ef\u4ee5\u66f4\u5feb\u5730\u8a13\u7df4 GPT-3 175B \u7b49\u5927\u898f\u6a21 AI \u6a21\u578b\uff0c\u9019\u7a2e\u51fa\u8272\u7684 AI \u6548\u80fd\u53ef\u4ee5\u8f49\u5316\u70ba\u5de8\u5927\u7684\u5546\u6a5f\u3002\u4f8b\u5982\uff0c\u5728 NVIDIA \u6700\u8fd1\u7684\u8ca1\u5831\u96fb\u8a71\u6703\u8b70\u4e2d\uff0c\u6211\u5011\u63cf\u8ff0\u4e86\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u670d\u52d9\u4f9b\u61c9\u5546\u5982\u4f55\u5728 NVIDIA HGX H200 \u4f3a\u670d\u5668\u4e0a\u904b\u884c Llama 3 70B \u6a21\u578b\uff0c\u5728\u77ed\u77ed\u56db\u5e74\u5167\u5c07\u4e00\u7f8e\u5143\u6295\u8cc7\u8f49\u5316\u70ba\u4e03\u7f8e\u5143\u3002\u9019\u500b\u6295\u8cc7\u56de\u5831\u662f\u5047\u8a2d\u4e00\u5bb6\u5927\u578b\u8a9e\u8a00\u670d\u52d9\u4f9b\u61c9\u5546\u4f7f\u7528\u541e\u5410\u91cf\u70ba\u6bcf\u79d2 24,000\u8a5e\u5143<\/a>\u7684 HGX H200 \u4f3a\u670d\u5668\uff0c\u4ee5\u6bcf\u767e\u842c\u8a5e\u5143 0.6 \u7f8e\u5143\u7684\u50f9\u683c\u63d0\u4f9b Llama 3 70B \u670d\u52d9\u3002<\/p>\n\n\n\n

\"\"
\u25b2\u64f4\u5c55\u5230\u65b0\u9ad8\u5ea6\uff1aNVIDIA MLPerf \u8a13\u7df4\u7d50\u679c\u5c55\u793a\u4e86\u524d\u6240\u672a\u6709\u7684\u6548\u80fd\u548c\u5f48\u6027<\/figcaption><\/figure>\n\n\n\n

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

NVIDIA H200 GPU\u589e\u5f37\u751f\u6210\u5f0fAI\u548cHPC<\/strong><\/h3>\n\n\n\n

NVIDIA H200 Tensor GPU \u57fa\u65bc Hopper \u67b6\u69cb\u7684\u512a\u52e2\u800c\u69cb\u5efa\uff0c\u64c1\u6709 141GB HBM3 \u8a18\u61b6\u9ad4\uff0c\u8207 H100 GPU \u76f8\u6bd4\uff0c\u8a18\u61b6\u9ad4\u983b\u5bec\u589e\u52a0\u4e86 40% \u4ee5\u4e0a\u3002 NVIDIA H200 Tensor Core GPU \u7a81\u7834\u4e86 AI \u8a13\u7df4\u7684\u6975\u9650\uff0c\u5728\u5176\u9996\u6b21\u4eae\u76f8\u7684 MLPerf Training \u4e2d\u5ef6\u4f38 H100 \u7684\u6548\u80fd\u4e26\u63d0\u9ad8\u4e86 47%\u3002<\/p>\n\n\n\n

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

\u8edf\u9ad4\u5e36\u4f86\u7121\u8207\u502b\u6bd4\u7684\u6548\u80fd\u63d0\u5347<\/strong><\/strong><\/h3>\n\n\n\n

\u6b64\u5916\uff0c\u7531\u65bc\u5c0d NVIDIA \u8edf\u9ad4\u5806\u758a\u9032\u884c\u4e86\u5927\u91cf\u6700\u4f73\u5316\uff0c\u6211\u5011\u4f7f\u7528 512 \u500b H100 GPU \u7684\u914d\u7f6e\u6240\u63d0\u4ea4\u7684\u7d50\u679c\u73fe\u5728\u6bd4\u4e00\u5e74\u524d\u5feb\u4e86 27%\u3002\u9019\u9805\u6539\u9032\u51f8\u986f\u4e86\u5373\u4f7f\u4f7f\u7528\u76f8\u540c\u7684\u786c\u9ad4\uff0c\u6301\u7e8c\u7684\u8edf\u9ad4\u589e\u5f37\u4e5f\u53ef\u4ee5\u986f\u8457\u63d0\u9ad8\u6548\u80fd\u3002<\/p>\n\n\n\n

\u9019\u9805\u5de5\u4f5c\u4e5f\u5be6\u73fe\u4e86\u8fd1\u4e4e\u5b8c\u7f8e\u7684\u64f4\u5145\u3002\u96a8\u8457 GPU \u6578\u91cf\u5f9e\u53bb\u5e74\u7684 3,584 \u500b H100 GPU \u589e\u52a0\u5230\u6b64\u6b21\u63d0\u4ea4\u7684 11,616 \u500b H100 GPU\uff0c\u589e\u52a0 3.2 \u500d\uff0c\u63d0\u4ea4\u7684\u6548\u80fd\u4e5f\u96a8\u4e4b\u7b49\u6bd4\u589e\u52a0\u3002<\/p>\n\n\n\n

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

\u5728\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08<\/strong>LLM<\/strong>\uff09\u5fae\u8abf\u65b9\u9762\u7684\u5353\u8d8a\u8868\u73fe<\/strong><\/strong><\/h3>\n\n\n\n

\u96a8\u8457\u4f01\u696d\u5c0b\u6c42\u5ba2\u88fd\u5316\u9810\u8a13\u7df4\u7684\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff0c\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u5fae\u8abf\u6b63\u5728\u6210\u70ba\u7522\u696d\u95dc\u9375\u7684\u5de5\u4f5c\u8ca0\u8f09\u3002 \u672c\u8f2a MLPerf \u5f15\u5165\u57fa\u65bc\u61c9\u7528\u65bc Meta Llama 2 70B \u7684\u71b1\u9580\u4f4e\u79e9\u9069\u61c9\uff08LoRA\uff09\u6280\u8853\u7684\u5168\u65b0\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u5fae\u8abf\u57fa\u6e96\u3002<\/p>\n\n\n\n

NVIDIA \u5e73\u53f0\u5728\u9019\u9805\u4efb\u52d9\u4e2d\u8868\u73fe\u51fa\u8272\uff0c\u5f9e 8 \u500b GPU \u64f4\u5c55\u5230 1,024 \u500b GPU\uff0cNVIDIA \u63d0\u4ea4\u4e86\u5728\u6700\u5927\u898f\u6a21\u7684\u904b\u7b97\u7d50\u679c\u5275\u7d00\u9304\u7684 1.5 \u5206\u9418\u5167\u5b8c\u6210\u4e86\u57fa\u6e96\u6e2c\u8a66<\/p>\n\n\n\n

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

\u52a0\u901fStable Diffusion\u548cGNN\u8a13\u7df4<\/strong><\/h3>\n\n\n\n

NVIDIA \u4e5f\u5728\u4e0a\u4e00\u8f2a\u63d0\u4ea4\u7684\u76f8\u540c\u7cfb\u7d71\u898f\u6a21\u4e0b\u5c07 Stable Diffusion v2 \u8a13\u7df4\u6548\u80fd\u63d0\u9ad8\u4e86 80%\u3002\u9019\u4e9b\u9032\u6b65\u53cd\u6620\u4e86 NVIDIA \u8edf\u9ad4\u5806\u758a\u7684\u8af8\u591a\u5f37\u5316\uff0c\u5c55\u793a\u4e86\u8edf\u9ad4\u548c\u786c\u9ad4\u6539\u9032\u5982\u4f55\u4e26\u9032\u4ee5\u63d0\u4f9b\u9802\u7d1a\u6548\u80fd\u3002<\/p>\n\n\n\n

\u5728\u57fa\u65bc R-GAT \u7684\u65b0\u5716\u795e\u7d93\u7db2\u8def\uff08GNN\uff09\u6e2c\u8a66\u4e2d\uff0c\u914d\u5099 H100 GPU \u7684 NVIDIA \u5e73\u53f0\u5728\u5c0f\u898f\u6a21\u548c\u5927\u898f\u6a21\u65b9\u9762\u5747\u8868\u73fe\u51fa\u8272\u3002\u8207 H100 \u76f8\u6bd4\uff0cH200 \u5728\u55ae\u7bc0\u9ede GNN \u8a13\u7df4\u65b9\u9762\u63d0\u5347\u4e86 47%\u3002\u9019\u5c55\u793a\u4e86 NVIDIA GPU \u7684\u5f37\u5927\u6548\u80fd\u548c\u9ad8\u6548\u7387\uff0c\u4f7f\u5176\u6210\u70ba\u5404\u7a2e AI \u61c9\u7528\u7684\u7406\u60f3\u9078\u64c7\u3002<\/p>\n\n\n\n

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

\u5ee3\u6cdb\u7684\u751f\u614b\u7cfb\u652f\u6301<\/strong><\/strong><\/h3>\n\n\n\n

10 \u5bb6 NVIDIA \u5408\u4f5c\u5925\u4f34\u63d0\u4ea4\u4e86\u7d50\u679c\uff0c\u53cd\u6620\u4e86 NVIDIA AI \u751f\u614b\u7cfb\u7684\u5ee3\u5ea6\uff0c\u5305\u62ec\u83ef\u78a9\u3001\u6234\u723e\u79d1\u6280\u96c6\u5718\u3001\u5bcc\u58eb\u901a\u3001\u6280\u5609\u79d1\u6280\u3001\u6167\u8207\u4f01\u696d\u3001\u806f\u60f3\u3001\u7532\u9aa8\u6587\u3001\u96f2\u9054\u79d1\u6280\u3001\u7f8e\u8d85\u5fae\u548c Sustainable Metal Cloud\u3002\u6b64\u5ee3\u6cdb\u7684\u53c3\u8207\u4ee5\u53ca\u5404\u5925\u4f34\u5091\u51fa\u7684\u57fa\u6e96\u6e2c\u8a66\u7d50\u679c\uff0c\u7a81\u986f\u4e86 NVIDIA AI \u5e73\u53f0\u5728\u6574\u500b\u7522\u696d\u7684\u5ee3\u6cdb\u63a1\u7528\u548c\u4fe1\u4efb\u3002<\/p>\n\n\n\n

MLCommons<\/a> \u6301\u7e8c\u81f4\u529b\u65bc\u5c07\u57fa\u6e96\u6e2c\u8a66\u6700\u4f73\u5be6\u8e10\u5f15\u5165 AI \u904b\u7b97\u81f3\u95dc\u91cd\u8981\u3002\u900f\u904e\u5c0d AI \u548c HPC \u5e73\u53f0\u9032\u884c\u540c\u5115\u5be9\u67e5\u6bd4\u8f03\uff0c\u4e26\u8ddf\u4e0a AI \u904b\u7b97\u7684\u5feb\u901f\u8b8a\u5316\uff0cMLCommons \u70ba\u4e16\u754c\u5404\u5730\u7684\u516c\u53f8\u63d0\u4f9b\u4e86\u6709\u52a9\u65bc\u5f15\u5c0e\u91cd\u8981\u63a1\u8cfc\u6c7a\u7b56\u7684\u95dc\u9375\u6578\u64da\u3002<\/p>\n\n\n\n

\u96a8\u8457 NVIDIA Blackwell<\/a> \u5e73\u53f0\u63a8\u51fa\uff0c\u7528\u65bc\u8a13\u7df4\u548c\u63a8\u8ad6\u7684\u5146\u53c3\u6578\u751f\u6210\u5f0f AI \u6a21\u578b\u7684\u65b0\u4e00\u7b49\u7d1a AI \u6548\u80fd\u5373\u5c07\u5be6\u73fe\u3002<\/p>\n\n\n\n

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

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

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

\ud83d\udfe6\u770b\u6587\u7ae0\u3001\u5f97\u597d\u79ae\uff1aSeasonic\u6d77\u97fb\u96fb\u5b50Computex 2024\u5c55\u5834\u5de1\u79aeFB\u6d3b\u52d5<\/a>
\u770b\u6587\u7ae0\u5f97\u597d\u79ae\uff1abe quiet! Computex 2024\u5c55\u5834\u5de1\u79ae FB\u6d3b\u52d5<\/a><\/strong>
\ud83d\udfe6\u66f4\u591a\u300aCOMPUTEX 2024\u300b<\/a>\u5831\u5c0e!!
\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\udfe6\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","protected":false},"excerpt":{"rendered":"

NVIDIA\u5728\u6700\u65b0\u7684MLPerf\u8a13\u7df4<\/p>\n","protected":false},"author":3,"featured_media":71625,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[580,6360,71],"class_list":["post-71605","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-focus","tag-mlperf","tag-nvidia"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/71605"}],"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=71605"}],"version-history":[{"count":2,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/71605\/revisions"}],"predecessor-version":[{"id":71626,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/71605\/revisions\/71626"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/71625"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=71605"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=71605"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=71605"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}