{"id":25684,"date":"2023-03-24T09:33:00","date_gmt":"2023-03-24T01:33:00","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=25684"},"modified":"2023-03-24T14:35:30","modified_gmt":"2023-03-24T06:35:30","slug":"nvidia-hopper-gpu%e4%b8%8d%e6%96%b7%e6%93%b4%e5%a4%a7%e6%87%89%e7%94%a8%e4%bb%a5%e6%94%af%e6%8f%b4%e5%b8%82%e5%a0%b4%e5%b0%8d%e4%ba%ba%e5%b7%a5%e6%99%ba%e6%85%a7%e9%9c%80%e6%b1%82%e7%9a%84%e5%a2%9e","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=25684","title":{"rendered":"NVIDIA Hopper GPU\u4e0d\u65b7\u64f4\u5927\u61c9\u7528\u4ee5\u652f\u63f4\u5e02\u5834\u5c0d\u4eba\u5de5\u667a\u6167\u9700\u6c42\u7684\u589e\u52a0"},"content":{"rendered":"\n<h3 class=\"wp-block-heading\">\u5404\u5927\u96f2\u7aef\u904b\u7b97\u670d\u52d9\u696d\u8005\u73fe\u5df2\u958b\u59cb\u63d0\u4f9bNVIDIA H100\u67b6\u69cb\uff0c\u4ee5\u6eff\u8db3\u5c0d\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u548c\u63a8\u8ad6\u670d\u52d9\u6301\u7e8c\u51fa\u73fe\u7684\u9700\u6c42\uff1bMeta\u3001OpenAI\u8207Stability.ai\u5c07\u5229\u7528H100\u4f86\u63a8\u52d5\u4e0b\u4e00\u6ce2\u4eba\u5de5\u667a\u6167\u6d6a\u6f6e<\/h3>\n\n\n\n<p>NVIDIA \u8207\u4e3b\u8981\u5408\u4f5c\u5925\u4f34\u4eca\u65e5\u5ba3\u5e03\u63a8\u51fa\u63a1\u7528 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/h100\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA H100 Tensor \u6838\u5fc3 GPU<\/a> \u7684\u65b0\u7522\u54c1\u548c\u670d\u52d9\uff0cNVIDIA H100 Tensor \u6838\u5fc3 GPU \u662f\u5168\u7403\u6700\u5f37\u5927\u7684\u4eba\u5de5\u667a\u6167 GPU\uff0c\u53ef\u6eff\u8db3\u5e02\u5834\u5c0d\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u548c\u63a8\u8ad6\u5feb\u901f\u51fa\u73fe\u7684\u9700\u6c42\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/blogs.oracle.com\/cloud-infrastructure\/post\/limited-availability-oci-compute-nvidia-h100\" target=\"_blank\" rel=\"noreferrer noopener\">Oracle Cloud Infrastructure<\/a> (OCI) \u5ba3\u5e03\u63a8\u51fa\u65b0\u7684 OCI \u904b\u7b97\u88f8\u6a5f GPU \u57f7\u884c\u500b\u9ad4\u5c07\u642d\u8f09 H100 GPU\uff0c\u63d0\u4f9b\u7684\u6578\u91cf\u6709\u9650\u3002\u6b64\u5916\uff0cAWS \u5ba3\u5e03\u5373\u5c07\u63a8\u51fa Amazon EC2 P5 \u57f7\u884c\u500b\u9ad4\u7684 EC2 UltraClusters\uff0c\u5176\u898f\u6a21\u53ef\u4ee5\u64f4\u5927\u5230\u5169\u842c\u500b\u4e92\u9023\u7684 H100 GPU\u3002\u9019\u662f\u7e7c<a href=\"https:\/\/azure.microsoft.com\/en-us\/blog\/azure-previews-powerful-and-scalable-virtual-machine-to-help-customers-accelerate-ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u5fae\u8edf Azure<\/a> \u4e0a\u9031\u5ba3\u5e03\u63a8\u51fa\u5176\u642d\u8f09 H100 \u865b\u64ec\u6a5f\u5668\u53ca ND H100 v5 \u5c08\u5c6c\u9810\u89bd\u6d3b\u52d5 (\u63d0\u4f9b\u5ba2\u6236\u6436\u5148\u9ad4\u9a57)\u4e4b\u5f8c\u7684\u53c8\u4e00\u9805\u91cd\u8981\u516c\u544a\u4e8b\u9805\u3002<\/p>\n\n\n\n<p>Meta \u73fe\u5df2\u5728\u516c\u53f8\u5167\u90e8\u90e8\u7f72\u4e86\u642d\u8f09 H100 \u7684 Grand Teton AI \u8d85\u7d1a\u96fb\u8166\uff0c\u4f9b\u4eba\u5de5\u667a\u6167\u751f\u7522\u548c\u7814\u7a76\u5718\u968a\u4f7f\u7528\u3002<\/p>\n\n\n\n<p>NVIDIA \u5275\u8fa6\u4eba\u66a8\u57f7\u884c\u9577\u9ec3\u4ec1\u52f3\u4eca\u65e5\u5728 GTC \u5927\u6703\u4e3b\u984c\u6f14\u8b1b\u4e2d\u5ba3\u5e03\uff0c<a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/h100\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA DGX H100<\/a> AI \u8d85\u7d1a\u96fb\u8166\u5df2\u5168\u9762\u6295\u7522\uff0c\u5f88\u5feb\u5c07\u4f9b\u5168\u7403\u4f01\u696d\u4f7f\u7528\u3002<\/p>\n\n\n\n<p>\u9ec3\u4ec1\u52f3\u8868\u793a\uff1a\u300c\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u7121\u6bd4\u7684\u767c\u5c55\u6f5b\u529b\uff0c\u6fc0\u52f5\u5e7e\u4e4e\u5404\u884c\u5404\u696d\u91cd\u65b0\u8a2d\u60f3\u5176\u5546\u696d\u7b56\u7565\u53ca\u5be6\u73fe\u9019\u4e9b\u7b56\u7565\u6240\u9700\u7684\u6280\u8853\u3002NVIDIA \u8207\u6211\u5011\u7684\u5408\u4f5c\u5925\u4f34\u5feb\u901f\u63a1\u53d6\u5404\u9805\u884c\u52d5\uff0c\u5c07\u4e16\u754c\u4e0a\u6700\u5f37\u5927\u7684\u4eba\u5de5\u667a\u6167\u904b\u7b97\u5e73\u53f0\u63d0\u4f9b\u7d66\u90a3\u4e9b\u958b\u767c\u5404\u7a2e\u61c9\u7528\u7a0b\u5f0f\u7684\u4eba\u54e1\uff0c\u800c\u9019\u4e9b\u61c9\u7528\u7a0b\u5f0f\u5c07\u5f9e\u6839\u672c\u4e0a\u6539\u8b8a\u6211\u5011\u751f\u6d3b\u3001\u5de5\u4f5c\u548c\u5a1b\u6a02\u7684\u65b9\u5f0f\u3002\u300d<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>Hopper <\/strong><strong>\u67b6\u69cb\u52a0\u901f\u63a8\u52d5\u4eba\u5de5\u667a\u6167\u767c\u5c55<\/strong><strong><\/strong><\/h3>\n\n\n\n<p>\u5efa\u7acb\u5728 NVIDIA Hopper GPU \u904b\u7b97\u67b6\u69cb\u4e0a\u7684 H100\uff0c\u5167\u5efa Transformer Engine\uff0c\u91dd\u5c0d\u958b\u767c\u3001\u8a13\u7df4\u548c\u90e8\u7f72\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u3001\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09\u53ca\u63a8\u85a6\u7cfb\u7d71\u9032\u884c\u4e86\u6700\u4f73\u5316\u8abf\u6574\u3002\u8207\u4e0a\u4e00\u4ee3 A100 \u76f8\u6bd4\uff0c\u9019\u9805\u6280\u8853\u5229\u7528\u4e86 H100 \u7684 FP8 \u7cbe\u5ea6\uff0c\u5728 LLM \u4e0a\u8a13\u7df4\u4eba\u5de5\u667a\u6167\u7684\u901f\u5ea6\u5feb\u4e86\u4e5d\u500d\u3001\u4eba\u5de5\u667a\u6167\u63a8\u8ad6\u901f\u5ea6\u5feb\u4e0a 30 \u500d\u3002H100 \u81ea\u79cb\u5b63\u8d77\u5c07\u96a8\u8457\u5168\u7403\u88fd\u9020\u5546\u7684\u500b\u5225\u548c\u7cbe\u9078\u4e3b\u6a5f\u677f\u55ae\u5143\u51fa\u8ca8\u3002<\/p>\n\n\n\n<p>NVIDIA DGX H100 \u642d\u8f09\u516b\u500b H100 GPU\uff0c\u4f7f\u7528 NVIDIA NVLink \u9ad8\u901f\u4e92\u9023\u6280\u8853\u53ca\u6574\u5408\u5f0f NVIDIA Quantum InfiniBand \u548c Spectrum Ethernet \u7db2\u8def\u6280\u8853\u9023\u63a5\u3002\u8a72\u5e73\u53f0\u4ee5 FP8 \u7cbe\u5ea6\u63d0\u4f9b 32 petaflops \u7684\u904b\u7b97\u6548\u80fd\uff0c\u7db2\u8def\u901f\u5ea6\u662f\u524d\u4e00\u4ee3 DGX A100 \u7684\u5169\u500d\uff0c\u6709\u52a9\u65bc\u5728\u8655\u7406\u5927\u578b\u4eba\u5de5\u667a\u6167\u5de5\u4f5c\u8ca0\u8f09\u6642\uff0c\u5c07\u80fd\u6e90\u4f7f\u7528\u6548\u7387\u63d0\u9ad8\u5230\u6975\u81f4\u3002<\/p>\n\n\n\n<p>DGX H100 \u4e5f\u5177\u5099\u5b8c\u6574\u7684 NVIDIA AI \u8edf\u9ad4\u5806\u758a\uff0c\u4f01\u696d\u80fd\u5920\u7528\u4ee5\u5927\u898f\u6a21\u5730\u6d41\u66a2\u57f7\u884c\u548c\u7ba1\u7406\u4eba\u5de5\u667a\u6167\u5de5\u4f5c\u8ca0\u8f09\u3002\u8a72\u7522\u54c1\u5305\u62ec\u4eca\u65e5\u53e6\u884c<a href=\"https:\/\/developer.nvidia.com\/blog\/supercharging-production-ai-everywhere-with-ai-enterprise-3-1\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u767c\u5e03<\/a>\u7684\u6700\u65b0\u7248 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/products\/ai-enterprise\/\" target=\"_blank\" rel=\"noopener\">NVIDIA AI E<\/a><a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/products\/ai-enterprise\/\" target=\"_blank\" rel=\"noreferrer noopener\">n<\/a><a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/products\/ai-enterprise\/\" target=\"_blank\" rel=\"noopener\">terprise<\/a>\uff0c\u4ee5\u53ca DGX \u8cc7\u6599\u4e2d\u5fc3\u7684\u4f5c\u696d\u7cfb\u7d71 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/base-command\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Base Command<\/a>\uff0c\u8a72\u7cfb\u7d71\u53ef\u4ee5\u5354\u8abf\u6574\u500b NVIDIA DGX \u5e73\u53f0\u4e0a\u7684\u4eba\u5de5\u667a\u6167\u8a13\u7df4\u548c\u64cd\u4f5c\u4e8b\u5b9c\uff0c\u8b93\u958b\u767c\u4eba\u5de5\u667a\u6167\u7684\u5de5\u4f5c\u8b8a\u5f97\u66f4\u7c21\u55ae\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"576\" src=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2023\/03\/20230324-nv10-01-1024x576.jpg\" alt=\"\" class=\"wp-image-25700\" title=\"\" srcset=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2023\/03\/20230324-nv10-01-1024x576.jpg 1024w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2023\/03\/20230324-nv10-01-300x169.jpg 300w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2023\/03\/20230324-nv10-01-768x432.jpg 768w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2023\/03\/20230324-nv10-01.jpg 1200w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u63a1\u7528<\/strong><strong> H100 <\/strong><strong>\u7684\u4eba\u5de5\u667a\u6167\u5148\u9a45<\/strong><strong><\/strong><\/h3>\n\n\n\n<p>\u591a\u500b\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u9818\u57df\u7684\u5148\u9a45\u63a1\u7528 H100 \u4f86\u52a0\u901f\u57f7\u884c\u5404\u9805\u5de5\u4f5c\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>OpenAI \u5148\u524d\u4f7f\u7528 H100 \u7684\u524d\u8eab &#8211; NVIDIA A100 GPU\uff0c\u4f86\u8a13\u7df4\u548c\u904b\u57f7\u884c\u5176 ChatGPT\uff0c\u9019\u662f\u4e00\u500b\u91dd\u5c0d\u5c0d\u8a71\u9032\u884c\u512a\u5316\u7684AI\u7cfb\u7d71\uff0c\u5728\u5275\u7d00\u9304\u7684\u6642\u9593\u88ab\u5168\u7403\u6578\u5104\u4eba\u4f7f\u7528\u3002OpenAI \u5c07\u5728\u5176 Azure \u8d85\u7d1a\u96fb\u8166\u4e0a\u4f7f\u7528 H100 \u4f86\u6301\u7e8c\u652f\u6301\u5176 AI \u7814\u7a76\u3002<\/li>\n\n\n\n<li>NVIDIA \u7684\u91cd\u8981\u6280\u8853\u5408\u4f5c\u5925\u4f34 <a href=\"https:\/\/nam11.safelinks.protection.outlook.com\/?url=https%3A%2F%2Fblogs.nvidia.com%2Fblog%2F2022%2F10%2F18%2Fmeta-grand-teton%2F&amp;data=05%7C01%7Ckbryson%40nvidia.com%7C0f5a7b8e3929480c672b08db27e943b5%7C43083d15727340c1b7db39efd9ccc17a%7C0%7C0%7C638147655038364023%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&amp;sdata=xvf7%2B9jiNTv6e6T6sFzRZKYrFXBeYGQfP7TazrMrlco%3D&amp;reserved=0\" target=\"_blank\" rel=\"noreferrer noopener\">Meta<\/a> \u958b\u767c\u642d\u8f09 Hopper \u7684\u4eba\u5de5\u667a\u6167\u8d85\u7d1a\u96fb\u8166 Grand Teton \u7cfb\u7d71\uff0c\u591a\u9805\u8868\u73fe\u512a\u65bc\u4e0a\u4e00\u4ee3\u7684 Zion \u7cfb\u7d71\uff0c\u5305\u62ec 4 \u500d\u7684\u4e3b\u6a5f\u5c0d GPU (host-to-GPU ) \u983b\u5bec\u30012 \u500d\u7684\u904b\u7b97\u548c\u6578\u64da\u7db2\u8def\u983b\u5bec\u4ee5\u53ca 2 \u500d\u7684\u5c01\u5305\u529f\u7387\u3002\u6709\u4e86\u9019\u7a2e\u66f4\u5f37\u5927\u7684\u904b\u7b97\u80fd\u529b\uff0cGrand Teton \u53ef\u4ee5\u652f\u6301\u9ad8\u7d1a\u6df1\u5ea6\u5b78\u7fd2\u63a8\u85a6\u6a21\u578b\u548c\u5167\u5bb9\u7406\u89e3\u7684\u8a13\u7df4\u548c\u751f\u7522\u63a8\u7406\u3002<\/li>\n\n\n\n<li>\u6587\u5b57\u8f49\u5f71\u50cf\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u9818\u57df\u7684\u5148\u9a45\u696d\u8005 Stability.ai\uff0c\u662f AWS \u5e73\u53f0\u4e0a\u6436\u5148\u8a66\u7528 H100 \u7684\u5ba2\u6236\u3002Stability.ai \u8a08\u756b\u4f7f\u7528 H100 \u4f86\u52a0\u901f\u57f7\u884c\u5176\u5373\u5c07\u63a8\u51fa\u7684\u5f71\u7247\u30013D \u548c\u591a\u6a21\u614b\u6a21\u578b\u3002<\/li>\n\n\n\n<li>Twelve Labs \u662f\u4e00\u500b\u8b93\u4f01\u696d\u548c\u958b\u767c\u4eba\u54e1\u80fd\u5920\u53d6\u5f97\u591a\u6a21\u5f0f\u8996\u8a0a\u7406\u89e3\u7684\u5e73\u53f0\uff0c\u5b83\u8a08\u5283\u5728 OCI \u8d85\u7d1a\u96fb\u8166\u4e0a\u4f7f\u7528 H100 \u57f7\u884c\u500b\u9ad4\uff0c\u4f7f\u8996\u8a0a\u66f4\u5373\u6642\u3001\u667a\u6167\u4e14\u66f4\u6613\u65bc\u88ab\u641c\u7d22\u3002<\/li>\n\n\n\n<li>\u5275\u4f5c\u51fa\u4eba\u5de5\u667a\u6167\u8f14\u52a9\u6545\u4e8b\u5beb\u4f5c\u548c\u6587\u5b57\u8f49\u5f71\u50cf\u5408\u6210 NovelAI \u61c9\u7528\u7a0b\u5f0f\u7684 Anlatan\uff0c\u4f7f\u7528 CoreWeave \u96f2\u7aef\u5e73\u53f0\u4e0a\u7684 H100 \u57f7\u884c\u500b\u9ad4\u4f86\u5efa\u7acb\u6a21\u578b\u53ca\u9032\u884c\u63a8\u8ad6\u4f5c\u696d\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u5168\u7403\u5404\u5730\u4f7f\u7528 DGX H100 \u7684\u5ba2\u6236<\/strong><\/h3>\n\n\n\n<p>\u5168\u7403\u5404\u5730\u7684\u5275\u65b0\u8005\u6536\u5230\u7b2c\u4e00\u6ce2 DGX H100 \u7cfb\u7d71\uff0c\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u65e5\u672c\u7684\u6578\u4f4d\u5ee3\u544a\u53ca\u7db2\u8def\u670d\u52d9\u516c\u53f8 <strong>CyberAgent<\/strong>\uff0c\u5145\u5206\u5229\u7528\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u8207 LLM \u6280\u8853\uff0c\u5efa\u7acb\u4eba\u5de5\u667a\u6167\u88fd\u4f5c\u51fa\u7684\u6578\u4f4d\u5ee3\u544a\u548c\u540d\u4eba\u6578\u4f4d\u5b7f\u751f\u5316\u8eab\u3002<\/li>\n\n\n\n<li><a href=\"https:\/\/www.jhuapl.edu\/about\" target=\"_blank\" rel=\"noreferrer noopener\">\u7d04\u7ff0\u970d\u666e\u91d1\u65af\u5927\u5b78\u61c9\u7528\u7269\u7406\u5be6\u9a57\u5ba4<\/a>\u662f\u7f8e\u570b\u6700\u5927\u7684\u5927\u5b78\u9644\u5c6c\u7814\u7a76\u4e2d\u5fc3\uff0c\u5c07\u4f7f\u7528 DGX H100 \u4f86\u8a13\u7df4 LLM\u3002<\/li>\n\n\n\n<li>\u4f4d\u65bc\u65af\u5fb7\u54e5\u723e\u6469\u7684<strong>\u745e\u5178<\/strong><strong> KTH <\/strong><strong>\u7687\u5bb6\u7406\u5de5\u5b78\u9662<\/strong>\u662f\u6b50\u6d32\u9802\u5c16\u7684\u7406\u5de5\u5927\u5b78\uff0c\u5c07\u4f7f\u7528 DGX H100 \u70ba\u9ad8\u7b49\u6559\u80b2\u5b78\u5b50\u63d0\u4f9b\u6700\u5148\u9032\u7684\u96fb\u8166\u79d1\u5b78\u8ab2\u7a0b\u3002<\/li>\n\n\n\n<li>\u65e5\u672c\u5927\u578b\u5546\u793e\u4e4b\u4e00\u7684<a href=\"https:\/\/blogs.nvidia.com\/blog\/2023\/03\/21\/generative-ai-supercomputer-pharmaceutical-industry\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u4e09\u4e95\u7269\u7522<\/a>\uff0c\u7d93\u71df\u80fd\u6e90\u3001\u5065\u5eb7\u3001IT \u53ca\u901a\u4fe1\u7b49\u591a\u9805\u696d\u52d9\uff0c\u6b63\u5728\u4f7f\u7528 DGX H100 \u4f86\u6253\u9020\u65e5\u672c\u9996\u500b\u7528\u65bc\u85e5\u7269\u767c\u73fe\u7684\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u8d85\u7d1a\u96fb\u8166\u3002<\/li>\n\n\n\n<li>\u5384\u74dc\u591a\u723e\u5927\u578b\u96fb\u4fe1\u5546 Telconet \u6b63\u5728\u5efa\u7acb\u667a\u6167\u5f71\u50cf\u5206\u6790\u6280\u8853\uff0c\u4ee5\u6253\u9020\u66f4\u5b89\u5168\u7684\u57ce\u5e02\uff0c\u540c\u6642\u70ba\u4f7f\u7528\u5404\u7a2e\u897f\u73ed\u7259\u8a9e\u65b9\u8a00\u7684\u5ba2\u6236\u63d0\u4f9b\u8a9e\u8a00\u670d\u52d9\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u751f\u614b\u7cfb\u7d71\u7684\u652f\u6301<\/strong><\/h3>\n\n\n\n<p><strong><em>Atlanan- Atlanan <\/em><\/strong><strong><em>\u57f7\u884c\u9577<\/em><\/strong><strong><em> Eren Do\u011fan<\/em><\/strong><\/p>\n\n\n\n<p>\u300c\u6211\u5011\u628a\u6240\u6709\u5fc3\u529b\u90fd\u653e\u5728\u4eba\u5de5\u667a\u6167\u5275\u65b0\u8207\u4eba\u5de5\u667a\u6167\u512a\u5148\u7684\u7522\u54c1\u4e0a\u3002NVIDIA H100 GPU \u662f\u6700\u5148\u9032\u7684\u6a5f\u5668\u5b78\u7fd2\u52a0\u901f\u5668\uff0c\u8b93\u6211\u5011\u5728\u6a5f\u5668\u5b78\u7fd2\u696d\u754c\u5167\u4f54\u76e1\u7af6\u722d\u512a\u52e2\uff0c\u5b83\u9069\u7528\u65bc\u5f9e\u6a21\u578b\u8a13\u7df4\u5230\u6a21\u578b\u63a8\u8ad6\u7684\u5404\u7a2e\u61c9\u7528\u9805\u76ee\u3002\u300d<\/p>\n\n\n\n<p><strong><em>AWS &#8211; AWS Amazon EC2 <\/em><\/strong><strong><em>\u526f\u7e3d\u88c1<\/em><\/strong><strong><em>David Brown<\/em><\/strong><\/p>\n\n\n\n<p>\u300cAWS \u548c NVIDIA \u5df2\u7d93\u5408\u4f5c\u8d85\u904e 12 \u5e74\uff0c\u63d0\u4f9b\u5927\u898f\u6a21\u3001\u7d93\u6fdf\u9ad8\u6548\u57fa\u65bc GPU \u7684\u89e3\u6c7a\u65b9\u6848\uff0c AWS \u5728\u63d0\u4f9b\u4ee5\u57fa\u65bcGPU\u7684\u57f7\u884c\u500b\u9ad4\u65b9\u9762\u64c1\u6709\u7121\u8207\u502b\u6bd4\u7684\u7d93\u9a57\uff0c\u4e26\u63a8\u52d5\u4e86\u6bcf\u4e00\u4ee3\u7522\u54c1\u7684\u53ef\u64f4\u5c55\u6027\u3002 \u5982\u4eca\uff0c\u8a31\u591a\u5ba2\u6236\u5c07\u6a5f\u5668\u5b78\u7fd2\u8a13\u7df4\u5de5\u4f5c\u8ca0\u8f09\u64f4\u5c55\u5230\u8d85\u904e 10,000 \u500b GPU\u3002 \u501f\u52a9\u7b2c\u4e8c\u4ee3 EFA\uff0c\u5ba2\u6236\u53ef\u4ee5\u5c07\u4ed6\u5011\u7684 P5 \u57f7\u884c\u500b\u9ad4\u64f4\u5c55\u5230\u8d85\u904e 20,000 \u500b H100 GPU\uff0c\u5f9e\u800c\u70ba\u4efb\u4f55\u7d44\u7e54\u5e36\u4f86\u7b26\u5408\u4e0d\u540c\u9700\u6c42\u7684\u8d85\u7d1a\u96fb\u8166\u529f\u80fd\u3002\u300d<\/p>\n\n\n\n<p><strong><em>Meta \u2013 Alexis Bjorlin<\/em><\/strong><strong><em>\uff0c\u57fa\u790e\u67b6\u69cb\u3001<\/em><\/strong><strong><em>AI <\/em><\/strong><strong><em>\u7cfb\u7d71\u548c\u52a0\u901f\u5e73\u53f0\u526f\u7e3d\u88c1<\/em><\/strong><strong><em><\/em><\/strong><\/p>\n\n\n\n<p>\u300c\u96a8\u8457\u6211\u5011\u5efa\u69cb\u65b0\u7684\u4eba\u5de5\u667a\u6167\u9ad4\u9a57\uff0c\u50cf\u662f\u57fa\u65bc\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u7684\u90a3\u4e9b\u9ad4\u9a57\uff0c\u5e95\u5c64\u7684\u4eba\u5de5\u667a\u6167\u6a21\u578b\u8b8a\u5f97\u8d8a\u4f86\u8d8a\u8907\u96dc\u3002Meta \u6700\u65b0\u7684\u63a1\u7528H100 \u9a45\u52d5\u7684 Grand Teton \u4eba\u5de5\u667a\u6167\u8d85\u7d1a\u96fb\u8166\u63d0\u4f9b\u4e86\u66f4\u5f37\u5927\u7684\u904b\u7b97\u80fd\u529b\u3001\u8a18\u61b6\u9ad4\u5bb9\u91cf\u548c\u983b\u5bec\uff0c\u9032\u4e00\u6b65\u52a0\u901f\u4e86 Meta \u7684 AI \u6a21\u578b\uff08\u4f8b\u5982\u958b\u6e90\u7684 DLRM\uff09\u7684\u8a13\u7df4\u548c\u63a8\u7406\u3002\u96a8\u8457\u6211\u5011\u958b\u767c\u51fa\u65b0\u4e00\u500b\u904b\u7b97\u5e73\u53f0\uff0c H100 \u4e5f\u70ba\u7814\u7a76 Meta \u672a\u4f86\u7684\u5167\u5bb9\u63a8\u85a6\u3001\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167 \u548c \u767c\u5c55\u5143\u5b87\u5b99\u9700\u6c42\u63d0\u4f9b\u66f4\u5f37\u5927\u7684\u904b\u7b97\u80fd\u529b\u3002\u300d<\/p>\n\n\n\n<p><strong><em>\u5fae\u8edf<\/em><\/strong><strong><em> \u2013 <\/em><\/strong><strong><em>\u5fae\u8edf\u96f2\u7aef\u904b\u7b97<\/em><\/strong><strong><em> + <\/em><\/strong><strong><em>\u4eba\u5de5\u667a\u6167\u4e8b\u696d\u7fa4\u57f7\u884c\u526f\u7e3d\u88c1<\/em><\/strong><strong><em> <\/em><\/strong><strong><em>Scott Guthrie<\/em><\/strong><\/p>\n\n\n\n<p>\u300c\u5404\u884c\u5404\u696d\u63a1\u7528\u4eba\u5de5\u667a\u6167\u7684\u901f\u5ea6\u4e0d\u65b7\u52a0\u5feb\uff0c\u4f01\u696d\u7d93\u71df\u8207\u53d6\u5f97\u6210\u529f\u7684\u65b9\u5f0f\u767c\u751f\u7ffb\u5929\u8986\u5730\u7684\u6539\u8b8a\u3002\u6211\u5011\u5728\u5fae\u8edf Azure \u5e73\u53f0\u4e0a\u4f7f\u7528 NVIDIA \u7684 Hopper \u67b6\u69cb\uff0c\u5c31\u80fd\u5c0d\u60f3\u8981\u64c1\u6709\u66f4\u591a\u4eba\u5de5\u667a\u6167\u80fd\u529b\u7684\u4f01\u696d\uff0c\u63d0\u4f9b\u7121\u8207\u502b\u6bd4\u7684\u904b\u7b97\u8868\u73fe\u548c\u529f\u80fd\u3002\u4eba\u5011\u63a1\u7528\u4eba\u5de5\u667a\u6167\u7684\u901f\u5ea6\u4e00\u76f4\u5728\u52a0\u5feb\uff0c\u4eba\u5de5\u667a\u6167\u6539\u8b8a\u4e86\u6bcf\u500b\u4f01\u696d\u7d93\u71df\u696d\u52d9\u8207\u53d6\u5f97\u6210\u529f\u7684\u65b9\u5f0f\u3002NVIDIA \u7684 Hopper \u67b6\u69cb\u70ba\u6108\u4f86\u6108\u591a\u4f7f\u7528\u5fae\u8edf Azure \u9032\u884c\u5275\u65b0\u7684\u4f01\u696d\uff0c\u5e36\u4f86\u524d\u6240\u672a\u6709\u7684\u904b\u7b97\u6548\u80fd\u3001\u898f\u6a21\u548c\u529f\u80fd\u3002\u300d<\/p>\n\n\n\n<p><strong><em>OpenAI<\/em><\/strong><strong><em>\u7684\u7e3d\u88c1\u66a8\u806f\u5408\u5275\u8fa6\u4eba<\/em><\/strong><strong><em>Greg Brockman<\/em><\/strong><\/p>\n\n\n\n<p>\u300cNVIDIA H100 Tensor Core GPU\u7684\u904b\u7b97\u80fd\u529b\u5c0d\u65bc\u63a8\u52d5AI\u8a13\u7df4\u548c\u63a8\u7406\u7684\u524d\u77bb\u767c\u5c55\u975e\u5e38\u91cd\u8981\u3002 NVIDIA\u7684\u5148\u9032\u6280\u8853\u5c07\u652f\u6301\u6211\u5011\u9032\u884c\u50cfGPT-4\u9019\u6a23\u7684\u7cfb\u7d71\u7814\u7a76\u548c\u5de5\u4f5c\u5354\u4f5c\u3002\u300d<\/p>\n\n\n\n<p><strong><em>Oracle Cloud Infrastructure &#8211; CTO<\/em><\/strong><strong><em>\u66a8\u9ad8\u7d1a\u526f\u7e3d\u88c1<\/em><\/strong><strong><em>Greg Pavlik<\/em><\/strong><\/p>\n\n\n\n<p>&nbsp;\u300cOCI\u6b63\u5728\u70ba\u6578\u5343\u5bb6\u5404\u7a2e\u898f\u6a21\u7684\u7d44\u7e54\u63d0\u4f9b\u5927\u898f\u6a21\u7684AI\u8d85\u7d1a\u904b\u7b97\u80fd\u529b\u3002\u6211\u5011\u8207NVIDIA\u7684\u7dca\u5bc6\u5408\u4f5c\u70ba\u5ba2\u6236\u63d0\u4f9b\u4e86\u5de8\u5927\u7684\u50f9\u503c\uff0c\u6211\u5011\u4e5f\u5c0d\u65bcH100\u7684\u5f37\u5927\u6548\u80fd\u611f\u5230\u975e\u5e38\u8208\u596e\u3002\u300d<\/p>\n\n\n\n<p><strong><em>Stability.ai &#8211; Stability.ai \u5275\u8fa6\u4eba\u66a8\u57f7\u884c\u9577 Emad Mostaque<\/em><\/strong><\/p>\n\n\n\n<p>Stability AI \u5275\u8fa6\u4eba\u66a8\u57f7\u884c\u9577 Emad Mostaque \u8868\u793a\uff1a\u300cStability AI \u662f\u9818\u5148\u5168\u7403\u7684\u958b\u653e\u539f\u59cb\u78bc\u4eba\u5de5\u667a\u6167\u6a21\u578b\u516c\u53f8\uff0c\u81f4\u529b\u65bc\u70ba\u6d88\u8cbb\u8005\u548c\u4f01\u696d\u63d0\u4f9b\u4e16\u754c\u4e0a\u6700\u4f73\u7684\u591a\u6a21\u614b\u5275\u4f5c\u5de5\u5177\u3002\u6211\u5011\u5229\u7528\u5f37\u5927\u7684 NVIDIA H100 \u63d0\u4f9b\u524d\u6240\u672a\u6709\u7684\u904b\u7b97\u80fd\u529b\uff0c\u70ba\u6108\u4f86\u6108\u591a\u60f3\u8981\u5f9e\u751f\u6210\u5f0f\u4eba\u5de5\u667a\u6167\u7684\u8b8a\u9769\u529b\u91cf\u4e2d\u53d7\u60e0\u7684\u4eba\u5011\uff0c\u652f\u6301\u4ed6\u5011\u767c\u63ee\u5275\u610f\u8207\u7814\u7a76\u80fd\u529b\u3002\u5b83\u5c07\u8b93\u6211\u5011\u88fd\u4f5c\u51fa\u66f4\u591a\u5f71\u7247\u30013D \u548c\u5176\u4ed6\u6a21\u578b\uff0c\u800c\u9019\u4e9b\u6a21\u578b\u5c07\u7368\u4eab\u66f4\u9ad8\u7684\u4e92\u9023\u901f\u5ea6\u53ca\u5148\u9032\u67b6\u69cb\u6240\u5e36\u4f86\u7684\u76ca\u8655\uff0c\u53ef\u7528\u65bc\u8655\u7406\u6d77\u91cf\u8cc7\u6599\u3002\u300d<\/p>\n\n\n\n<p><strong>Twelve Labs \u2013 <\/strong><strong>\u57f7\u884c\u9577<\/strong><strong> Jae Lee<\/strong><\/p>\n\n\n\n<p>\u300c\u6211\u5011\u5f88\u9ad8\u8208\u80fd\u5920\u5229\u7528\u642d\u8f09NVIDIA H100 GPU\u7684Oracle Cloud Infrastructure Compute\u88f8\u6a5f\u57f7\u884c\u500b\u9ad4\uff0c\u7e7c\u7e8c\u52aa\u529b\u5c07\u8996\u8a0a\u57fa\u790e\u6a21\u578b\u5c0e\u5165\u5e02\u5834\u3002\u300d<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u4e0a\u5e02\u6642\u9593<\/strong><\/h3>\n\n\n\n<p>NVIDIA DGX H100 \u8d85\u7d1a\u96fb\u8166\u5df2\u5168\u9762\u6295\u7522\uff0c\u53ef\u5411 NVIDIA \u5168\u7403\u5408\u4f5c\u5925\u4f34\u8a02\u8cfc\u3002\u5ba2\u6236\u73fe\u5728\u4fbf\u80fd\u900f\u904e <a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/dgx-cloud\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA DGX Cloud<\/a> \u9ad4\u9a57 DGX H100\u3002\u8acb\u5411\u5168\u7403 NVIDIA DGX \u5408\u4f5c\u5925\u4f34\u8a62\u554f\u50f9\u683c\u3002<\/p>\n\n\n\n<p>\u5728\u96f2\u7aef\u670d\u52d9\u696d\u8005\u65b9\u9762\uff0c\u5fae\u8edfAzure\u76ee\u524d\u63d0\u4f9b NVIDIA H100 \u5c08\u5c6c\u9810\u89bd\u670d\u52d9\uff0c\u63d0\u4f9b\u5ba2\u6236\u6436\u5148\u9ad4\u9a57\uff0c Oracle Cloud Infrastructure \u63d0\u4f9b\u6709\u9650\u7684\u4f7f\u7528\u3002<a href=\"https:\/\/cirrascale.com\/press-nvidia-h100.php\" target=\"_blank\" rel=\"noreferrer noopener\">Cirrascale<\/a><a href=\"https:\/\/cirrascale.com\/press-nvidia-h100.php\" target=\"_blank\" rel=\"noopener\"> <\/a>\u8207 <a href=\"https:\/\/www.coreweave.com\/blog\" target=\"_blank\" rel=\"noreferrer noopener\">CoreWeave<\/a>\u4e5f\u5df2\u63d0\u4f9b\u5177 H100\u7684\u96f2\u7aef\u5e73\u53f0\u3002AWS\u5ba3\u5e03\u5c07\u5728\u6578\u9031\u4e4b\u5f8c\u63a8\u51fa\u5c08\u5c6c\u9810\u89bd\u670d\u52d9\u3002Google\u8207Lambda\u3001Paperspace \u53ca Vultr\u7b49\u96f2\u7aef\u670d\u52d9\u5408\u4f5c\u5925\u4f34\u7686\u8a08\u756b\u63a8\u51fa\u642d\u8f09H100\u7684\u96f2\u7aef\u5e73\u53f0\u3002<\/p>\n\n\n\n<p>\u5404\u5927\u4f3a\u670d\u5668\u88fd\u9020\u5546\u73fe\u5df2\u958b\u59cb\u63d0\u4f9b\u642d\u8f09 NVIDIA H100 GPU \u7684\u4f3a\u670d\u5668\u548c\u7cfb\u7d71\uff0c\u9019\u4e9b\u696d\u8005\u6709\u6e90\u8a0a\uff08Atos\uff09 \u3001\u601d\u79d1\u7cfb\u7d71\uff08Cisco\uff09 \u3001\u6234\u723e\u79d1\u6280\u96c6\u5718\uff08Dell Technologies\uff09\u3001\u6280\u5609\u79d1\u6280\u3001\u6167\u8207\u79d1\u6280\u516c\u53f8\uff08Hewlett Packard Enterprise\uff09\u3001\u806f\u60f3\u96c6\u5718 \uff08Lenovo\uff09\u53ca \u7f8e\u8d85\u5fae\u96fb\u8166 \uff08Supermicro\uff09\u3002<\/p>\n\n\n\n<p>\u8acb\u76f4\u63a5\u5411 NVIDIA \u5408\u4f5c\u5925\u4f34\u8a62\u554f\u50f9\u683c\u8207\u5176\u4ed6\u8a73\u7d30\u8cc7\u8a0a\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h4 class=\"wp-block-heading has-text-align-right has-very-light-gray-to-cyan-bluish-gray-gradient-background has-background\"><br><strong><img loading=\"lazy\" decoding=\"async\" height=\"16\" width=\"16\" src=\"https:\/\/static.xx.fbcdn.net\/images\/emoji.php\/v9\/t83\/1\/16\/1f7e9.png\" alt=\"\ud83d\udfe9\" title=\"\"><\/strong>\u5beb\u51fa\u60a8\u7684\u71b1\u8840<strong>\u96fb\u8166\u7d93\u9a57\u5c31\u6709\u6a5f\u6703\u8b8a\u6587\u7ae0\u4e3b\u89d2 \u518d\u62bd\u8d85\u7d1a\u597d\u79ae<\/strong> <a href=\"https:\/\/www.facebook.com\/ioioTIMES\/posts\/pfbid0RKM1kpqM6HV6qH3rac3uznbeP3j9oXUWCRymwzxz2CyGYJErnwLmjgtN19PYarcHl\" target=\"_blank\" rel=\"noopener\">\u6d3b\u52d5\u7531\u6b64\u53bb<\/a><br>\ud83d\udfe6<strong>\u73fe\u5728\u5c31\u52a0\u5165&nbsp;<a href=\"https:\/\/www.facebook.com\/profile.php?id=100086628162118\" target=\"_blank\" rel=\"noreferrer noopener\">ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718<\/a>&nbsp;\u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!<\/strong><br><strong><img loading=\"lazy\" decoding=\"async\" height=\"16\" width=\"16\" src=\"https:\/\/static.xx.fbcdn.net\/images\/emoji.php\/v9\/t83\/1\/16\/1f7e9.png\" alt=\"\ud83d\udfe9\" title=\"\">\u6211\u5011\u6709<a href=\"https:\/\/today.line.me\/tw\/v2\/publisher\/103117\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">LINE TODAY<\/a>\u983b\u9053\u4e86\uff0c\u5feb\u4f86\u8ffd\u8e2a\u6211\u5011\u5427!!&#8211;\u6700\u65b0\u79d1\u6280\u65b0\u805e \u76e1\u5728\u4f60\u624b<\/strong><\/h4>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u5404\u5927\u96f2\u7aef\u904b\u7b97\u670d\u52d9\u696d\u8005\u73fe\u5df2\u958b\u59cb\u63d0\u4f9bNV<\/p>\n","protected":false},"author":3,"featured_media":25700,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[5360,3022,71,5361,5362,5363,642],"class_list":["post-25684","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-hopper-gpu","tag-meta","tag-nvidia","tag-nvidia-h100","tag-openai","tag-stability-ai","tag-642"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/25684"}],"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=25684"}],"version-history":[{"count":3,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/25684\/revisions"}],"predecessor-version":[{"id":25702,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/25684\/revisions\/25702"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/25700"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=25684"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=25684"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=25684"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}