{"id":111007,"date":"2025-06-05T11:46:20","date_gmt":"2025-06-05T03:46:20","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=111007"},"modified":"2025-06-05T11:46:21","modified_gmt":"2025-06-05T03:46:21","slug":"nvidia-blackwell%e5%9c%a8%e6%9c%80%e6%96%b0mlperf-training%e6%b8%ac%e8%a9%a6%e7%b5%90%e6%9e%9c%e4%b8%ad%e5%b1%95%e7%8f%be%e7%aa%81%e7%a0%b4%e6%80%a7%e6%95%88%e8%83%bd","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=111007","title":{"rendered":"NVIDIA Blackwell\u5728\u6700\u65b0MLPerf Training\u6e2c\u8a66\u7d50\u679c\u4e2d\u5c55\u73fe\u7a81\u7834\u6027\u6548\u80fd"},"content":{"rendered":"\n

NVIDIA \u6b63\u8207\u5168\u7403\u4f01\u696d\u4e26\u80a9\u5efa\u9020\u4eba\u5de5\u667a\u6167\uff08AI\uff09\u5de5\u5ee0<\/a>\uff0c\u52a0\u5feb\u8a13\u7df4\u53ca\u90e8\u7f72\u63a1\u7528\u6700\u65b0\u8a13\u7df4\u8207\u63a8\u8ad6\u6280\u8853\u7684\u4e0b\u4e00\u4ee3 AI \u61c9\u7528\u3002<\/p>\n\n\n\n

NVIDIA \u7684 Blackwell \u67b6\u69cb\u5c08\u70ba\u6eff\u8db3\u9019\u4e9b\u5c0d\u6548\u80fd\u9700\u6c42\u65e5\u6f38\u5347\u9ad8\u7684\u65b0\u578b\u61c9\u7528\u800c\u6253\u9020\u3002\u5728\u6700\u65b0\u4e00\u8f2a MLPerf Training \u6e2c\u8a66\u4e2d\uff0c\u540c\u6642\u4e5f\u662f\u81ea 2018 \u5e74\u63a8\u51fa\u9019\u9805\u57fa\u6e96\u6e2c\u8a66\u4ee5\u4f86\u7684\u7b2c 12 \u8f2a\u6e2c\u8a66\uff0cNVIDIA AI \u5e73\u53f0\u5728\u6bcf\u9805\u57fa\u6e96\u6e2c\u8a66\u4e2d\u5747\u5c55\u73fe\u51fa\u6700\u5927\u898f\u6a21\u6548\u80fd\uff0c\u4e26\u9a45\u52d5\u8a72\u57fa\u6e96\u6e2c\u8a66\u4e2d\u6700\u5177\u6311\u6230\u6027\u7684\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08LLM<\/a>\uff09\u6e2c\u8a66\u9805\u76ee\u300cLlama 3.1 405B \u9810\u8a13\u7df4\u300d\u7684\u6240\u6709\u7d50\u679c\u3002<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

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

NVIDIA \u5e73\u53f0\u662f\u552f\u4e00\u5728\u6bcf\u9805 MLPerf Training v5.0 \u57fa\u6e96\u6e2c\u8a66\u4e2d\u5747\u6709\u63d0\u4ea4\u7d50\u679c\u7684\u5e73\u53f0\uff0c\u9019\u7a81\u986f\u5176\u5728\u5404\u7a2e AI \u5de5\u4f5c\u8ca0\u8f09\u4e2d\u7684\u5353\u8d8a\u6548\u80fd\u8207\u4efb\u52d9\u591a\u6a23\u6027\uff0c\u5305\u62ec\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u3001\u63a8\u85a6\u7cfb\u7d71\u3001\u591a\u6a21\u614b\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u3001\u7269\u9ad4\u5075\u6e2c\u8207\u5716\u795e\u7d93\u7db2\u8def\uff08GNN\uff09<\/a>\u3002<\/p>\n\n\n\n

\u9019\u6b21\u5927\u898f\u6a21\u63d0\u4ea4\u7684\u6e2c\u8a66\u7d50\u679c\u4f7f\u7528\u5169\u90e8\u642d\u8f09 NVIDIA Blackwell \u5e73\u53f0\u7684 AI \u8d85\u7d1a\u96fb\u8166\uff1aTyche \u63a1\u7528 NVIDIA GB200 NVL72 <\/a>\u6a5f\u67b6\u898f\u6a21\u7cfb\u7d71\u6253\u9020\uff0cNyx \u5247\u4ee5 NVIDIA DGX B200<\/a> \u7cfb\u7d71\u70ba\u57fa\u790e\u3002\u6b64\u5916\uff0cNVIDIA \u4e5f\u8207 CoreWeave \u548c IBM \u5408\u4f5c\uff0c\u63d0\u4ea4 GB200 NVL72 \u7684\u6e2c\u8a66\u7d50\u679c\uff0c\u5171\u4f7f\u7528 2,496 \u9846 Blackwell GPU \u8207 1,248 \u9846 NVIDIA Grace CPU\u3002<\/p>\n\n\n\n

\u5728\u6700\u65b0\u7684 Llama 3.1 405B \u9810\u8a13\u7df4\u57fa\u6e96\u6e2c\u8a66\u4e2d\uff0cBlackwell \u7684\u6548\u80fd\u8f03\u4e0a\u4e00\u4ee3\u67b6\u69cb\u5728\u76f8\u540c\u898f\u6a21\u4e0b\u9ad8\u51fa 2.2 \u500d\u3002<\/p>\n\n\n\n

\u5728 Llama 2 70B LoRA \u5fae\u8abf\u57fa\u6e96\u6e2c\u8a66\u4e2d\uff0c\u642d\u8f09 8 \u9846 Blackwell GPU \u7684 NVIDIA DGX B200 \u7cfb\u7d71\u6548\u80fd\uff0c\u8f03\u4e0a\u4e00\u8f2a\u4f7f\u7528\u76f8\u540c\u6578\u91cf GPU \u7684\u6e2c\u8a66\u7d50\u679c\u63d0\u5347\u9054 2.5 \u500d\u3002<\/p>\n\n\n\n

\u9019\u4e9b\u6548\u80fd\u4e0a\u7684\u8e8d\u9032\u7a81\u986f Blackwell \u67b6\u69cb\u7684\u512a\u52e2\uff0c\u5305\u62ec\u9ad8\u5bc6\u5ea6\u6db2\u51b7\u6a5f\u67b6\u3001\u6bcf\u500b\u6a5f\u67b6\u642d\u914d 13.4TB \u7684\u4e00\u81f4\u6027\u8a18\u61b6\u9ad4\u3001\u7528\u65bc\u5782\u76f4\u64f4\u5c55\u7684\u7b2c\u4e94\u4ee3 NVIDIA NVLink \u53ca NVIDIA NVLink Switch<\/a> \u4e92\u9023\u6280\u8853\uff0c\u4ee5\u53ca\u7528\u65bc\u6a6b\u5411\u64f4\u5c55\u7684 NVIDIA Quantum-2 InfiniBand<\/a> \u7db2\u8def\u6280\u8853\u3002\u6b64\u5916\uff0cNVIDIA NeMo Framework<\/a> \u8edf\u9ad4\u5806\u758a\u7684\u5275\u65b0\u6280\u8853\uff0c\u63d0\u5347\u4e86\u4e0b\u4e00\u4ee3\u591a\u6a21\u614b\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u8a13\u7df4\u7684\u6c34\u6e96\uff0c\u9019\u5c0d\u65bc\u5c07\u4ee3\u7406\u578b AI<\/a> \u61c9\u7528\u63a8\u5411\u5e02\u5834\u800c\u8a00\u81f3\u95dc\u91cd\u8981\u3002<\/p>\n\n\n\n

\u9019\u4e9b\u7531\u4ee3\u7406\u578b AI \u9a45\u52d5\u7684\u61c9\u7528\uff0c\u6709\u671d\u4e00\u65e5\u5c07\u5728 AI \u5de5\u5ee0\u4e2d\u904b\u884c\uff0c\u800c\u9019\u4e9b AI \u5de5\u5ee0\u6b63\u662f\u63a8\u52d5\u4ee3\u7406\u578b AI \u7d93\u6fdf\u7684\u5f15\u64ce\u3002\u9019\u4e9b\u5168\u65b0\u61c9\u7528\u5c07\u7522\u751f\u5404\u7a2e\u8a5e\u5143\uff08token\uff09<\/a>\u8207\u5bf6\u8cb4\u7684\u667a\u6167\uff0c\u53ef\u61c9\u7528\u65bc\u5e7e\u4e4e\u6240\u6709\u7522\u696d\u8207\u5b78\u8853\u9818\u57df\u3002<\/p>\n\n\n\n

NVIDIA \u8cc7\u6599\u4e2d\u5fc3\u5e73\u53f0\u5305\u542b GPU\u3001CPU\u3001\u9ad8\u901f\u4e92\u9023\u67b6\u69cb\u548c\u7db2\u8def\u6280\u8853\uff0c\u4ee5\u53ca\u7a2e\u985e\u8c50\u5bcc\u7684\u8edf\u9ad4\uff0c\u5305\u62ec NVIDIA CUDA-X<\/a> \u51fd\u5f0f\u5eab\u3001NeMo Framework\u3001NVIDIA TensorRT-LLM<\/a> \u8207 NVIDIA Dynamo<\/a>\u3002\u9019\u5957\u9ad8\u5ea6\u6700\u4f73\u5316\u7684\u786c\u9ad4\u8207\u8edf\u9ad4\uff0c\u4f7f\u5f97\u7d44\u7e54\u80fd\u66f4\u5feb\u901f\u5730\u8a13\u7df4\u53ca\u90e8\u7f72\u6a21\u578b\uff0c\u986f\u8457\u52a0\u5feb\u5275\u9020\u50f9\u503c\u7684\u6642\u9593\u3002<\/p>\n\n\n\n

NVIDIA \u7684\u5408\u4f5c\u5925\u4f34\u751f\u614b\u7cfb\u5ee3\u6cdb\u53c3\u8207\u9019\u8f2a MLPerf \u6e2c\u8a66\u3002\u9664\u4e86\u8207 CoreWeave \u548c IBM \u5408\u4f5c\u63d0\u4ea4\u7684\u6e2c\u8a66\u7d50\u679c\u5916\uff0c\u83ef\u78a9<\/a>\u3001\u601d\u79d1\u3001\u6234\u723e\u79d1\u6280\u96c6\u5718\u3001\u6280\u92fc\u79d1\u6280\u3001Google Cloud\u3001\u6167\u8207\u79d1\u6280\u3001Lambda\u3001\u806f\u60f3\u96c6\u5718\u3001Nebius\u3001Oracle Cloud Infrastructure\u3001\u96f2\u9054\u79d1\u6280\u548c\u7f8e\u8d85\u5fae\u4ea6\u4ea4\u51fa\u4eae\u773c\u7684\u6e2c\u8a66\u7d50\u679c\u3002<\/p>\n\n\n\n

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

\u9032\u4e00\u6b65\u4e86\u89e3<\/em> <\/em>MLPerf <\/em>\u57fa\u6e96\u6e2c\u8a66<\/em><\/a>\u3002<\/em><\/p>\n\n\n\n

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

\u3000\u3000\u3000\u3000\u3000<\/p>\n\n\n\n

\ud83d\udfe6COMPUTEX \u6d3b\u52d5 \u2013 \u770b\u5341\u9293\u5c55\u5834\u4ecb\u7d39\u9001\u76ae\u9769\u5361\u5957\u5305<\/a>
\ud83d\udfe6COMPUTEX \u6d3b\u52d5 \u2013 \u770b ZOTAC \u5c55\u5834\u65b0\u54c1\u62ff\u9650\u91cf\u5c0f\u591c\u71c8<\/a>
\ud83d\udfe6COMPUTEX \u6d3b\u52d5– \u770b MSI \u5c55\u524d\u6703\u62ff\u8d85\u840c\u597d\u79ae<\/a>
\ud83d\udfe6\u641e\u754f\u6709\u79ae\uff01\u53ea\u8981\u65bc\u7ad9\u4e0a FB \u4efb\u4e00\u5e16\u767c\u8a00<\/a>\u5c31\u6709\u6a5f\u6703\u7372\u5f97\u300c\u641e\u754f\u734e\u795e\u79d8\u79ae\u54c1\u300d
\ud83d\udfe6\u73fe\u5728\u5c31\u52a0\u5165 ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718 \u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!
\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<\/strong>\u624b<\/h4>\n","protected":false},"excerpt":{"rendered":"

NVIDIA \u6b63\u8207\u5168\u7403\u4f01\u696d<\/p>\n","protected":false},"author":3,"featured_media":111010,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[580,13766,71],"class_list":["post-111007","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-focus","tag-mlperf-training","tag-nvidia"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/111007"}],"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=111007"}],"version-history":[{"count":1,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/111007\/revisions"}],"predecessor-version":[{"id":111012,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/111007\/revisions\/111012"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/111010"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=111007"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=111007"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=111007"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}