{"id":64155,"date":"2024-04-10T09:30:29","date_gmt":"2024-04-10T01:30:29","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=64155"},"modified":"2024-04-10T16:56:43","modified_gmt":"2024-04-10T08:56:43","slug":"%e8%8b%b1%e7%89%b9%e7%88%be%e6%89%93%e7%a0%b4%e5%b0%88%e5%88%a9%e9%99%90%e5%88%b6%ef%bc%8c%e7%82%ba%e4%bc%81%e6%a5%ad%e7%94%9f%e6%88%90%e5%bc%8fai%e5%b8%82%e5%a0%b4%e6%8f%90%e4%be%9b%e6%96%b0%e9%81%b8","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=64155","title":{"rendered":"\u82f1\u7279\u723e\u6253\u7834\u4f7f\u7528\u9650\u5236\uff0c\u70ba\u4f01\u696d\u751f\u6210\u5f0fAI\u5e02\u5834\u63d0\u4f9b\u65b0\u9078\u64c7"},"content":{"rendered":"\n<h3 class=\"wp-block-heading\">\u5ef6\u7e8cGaudi 2\u7684\u6548\u80fd\u548c\u53ef\u64f4\u5145\u6027\uff0cIntel Gaudi 3 AI\u52a0\u901f\u5668\u70ba\u5168\u7403\u4f01\u696d\u63d0\u4f9b\u751f\u6210\u5f0fAI\u65b0\u9078\u64c7<\/h3>\n\n\n\n<p>\u82f1\u7279\u723e\u5728 Vision 2024 \u5927\u6703\u4e0a\uff0c\u5ba3\u5e03\u63a8\u51fa Intel\u00ae Gaudi\u00ae 3 AI \u52a0\u901f\u5668\uff0c\u8207\u524d\u4ee3\u7522\u54c1\u76f8\u6bd4\uff0cGaudi 3 \u70ba BF16 \u63d0\u4f9b 4 \u500d AI \u904b\u7b97\u80fd\u529b\u30011.5 \u500d\u8a18\u61b6\u9ad4\u983b\u5bec\u4ee5\u53ca 2 \u500d\u7db2\u8def\u983b\u5bec\uff0c\u53ef\u64f4\u5145\u5927\u898f\u6a21\u7cfb\u7d71\uff0c\u5c07\u6709\u52a9\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09\u548c\u591a\u6a21\u614b\u6a21\u578b\u7684 AI \u8a13\u7df4\u548c\u63a8\u7406\uff0c\u5927\u5e45\u63d0\u5347\u6548\u80fd\u548c\u751f\u7522\u529b\u3002Intel\u00ae Gaudi\u00ae 2 AI \u52a0\u901f\u5668\u662f\u5e02\u5834\u4e0a<a href=\"https:\/\/www.intel.com\/content\/www\/us\/en\/newsroom\/news\/new-gaudi-2-xeon-performance-ai-inference.html\" target=\"_blank\" rel=\"noreferrer noopener\">\u552f\u4e00\u901a\u904eMLPerf\u57fa\u6e96\u6e2c\u8a66\u7684LLM\u89e3\u6c7a\u65b9\u6848<\/a>\uff0c\u6548\u80fd\u548c\u6548\u7387\u7686\u901a\u904e\u9a57\u8b49\u3002\u82f1\u7279\u723e\u900f\u904e\u958b\u6e90\u793e\u7fa4\u8edf\u9ad4\u548c\u7b26\u5408\u696d\u754c\u6a19\u6e96\u7684\u4e59\u592a\u7db2\u7d61\uff0c\u70ba\u5ba2\u6236\u63d0\u4f9b\u53ef\u9748\u6d3b\u64f4\u5145\u7cfb\u7d71\u7684\u65b0\u9078\u64c7\u3002<\/p>\n\n\n\n<p>\u82f1\u7279\u723e\u57f7\u884c\u526f\u7e3d\u88c1\u66a8\u8cc7\u6599\u4e2d\u5fc3\u8207 AI \u89e3\u6c7a\u65b9\u6848\u7e3d\u7d93\u7406 Justin Hotard \u8868\u793a\uff1a\u300cAI \u5e02\u5834\u77ac\u606f\u842c\u8b8a\uff0c\u4f46\u7522\u54c1\u9593\u4ecd\u5b58\u5728\u5de8\u5927\u5dee\u8ddd\u3002\u4e0d\u8ad6\u662f\u4f86\u81ea\u5ba2\u6236\u9084\u662f\u66f4\u5ee3\u6cdb\u5e02\u5834\u7684\u56de\u994b\uff0c\u7686\u53cd\u6620\u5c0d\u66f4\u591a\u9078\u64c7\u7684\u6e34\u671b\u3002\u4f01\u696d\u9808\u6b0a\u8861\u53ef\u7528\u6027\u3001\u53ef\u64f4\u5145\u6027\u3001\u6548\u80fd\u3001\u6210\u672c\u548c\u80fd\u6e90\u6548\u7387\u7b49\u56e0\u7d20\u3002Intel Gaudi 3 \u4f5c\u70ba\u751f\u6210\u5f0f AI \u7684\u65b0\u9078\u64c7\uff0c\u6191\u85c9\u6027\u50f9\u6bd4\u3001\u7cfb\u7d71\u53ef\u64f4\u5145\u6027\u548c\u6642\u9593\u6210\u672c\u512a\u52e2\u7684\u5b8c\u7f8e\u7d50\u5408\u812b\u7a4e\u800c\u51fa\u3002\u300d<\/p>\n\n\n\n<p>\u91d1\u878d\u3001\u88fd\u9020\u548c\u91ab\u7642\u4fdd\u5065\u7b49\u95dc\u9375\u9818\u57df\u7684\u4f01\u696d\uff0c\u76ee\u524d\u6b63\u5feb\u901f\u63d0\u5347 AI \u7684\u666e\u53ca\u5316\uff0c\u4e26\u7a4d\u6975\u5c07\u751f\u6210\u5f0f AI \u8a08\u756b\u5f9e\u8a66\u9a57\u968e\u6bb5\u8f49\u70ba\u5168\u9762\u5be6\u65bd\u3002\u70ba\u4e86\u56e0\u61c9\u8f49\u578b\u3001\u63a8\u52d5\u5275\u65b0\u4e26\u9054\u6210\u71df\u6536\u6210\u9577\u76ee\u6a19\uff0c\u4f01\u696d\u9700\u8981\u958b\u653e\u3001\u7b26\u5408\u6210\u672c\u6548\u76ca\u4e14\u66f4\u7bc0\u80fd\u7684\u89e3\u6c7a\u65b9\u6848\u548c\u7522\u54c1\uff0c\u4ee5\u7b26\u5408\u6295\u8cc7\u5831\u916c\u7387\uff08ROI\uff09\u548c\u71df\u904b\u6548\u7387\u9700\u6c42\u3002<\/p>\n\n\n\n<p>Intel Gaudi 3 \u52a0\u901f\u5668\u5c07\u6eff\u8db3\u9019\u4e9b\u9700\u6c42\uff0c\u4e26\u900f\u904e\u958b\u653e\u793e\u7fa4\u8edf\u9ad4\u548c\u958b\u653e\u6a19\u6e96\u7684\u4e59\u592a\u7db2\u8def\uff0c\u5354\u52a9\u4f01\u696d\u9748\u6d3b\u64f4\u5145 AI \u7cfb\u7d71\u548c\u61c9\u7528\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"661\" src=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240410-intel02.jpg\" alt=\"\" class=\"wp-image-64166\" title=\"\" srcset=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240410-intel02.jpg 1024w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240410-intel02-300x194.jpg 300w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2024\/04\/20240410-intel02-768x496.jpg 768w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">\u25b2Intel tackles the generative AI gap by introducing the Intel Gaudi 3 AI accelerator at the Intel Vision event on April 9, 2024, in Phoenix, Arizona. Gaudi 3 gives customers choice with open community-based software and industry-standard Ethernet networking to scale their systems more flexibly. (Credit: Intel Corporation)<\/figcaption><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p><\/p>\n\n\n\n<p><strong>\u5ba2\u88fd\u5316\u57fa\u790e\u67b6\u69cb\u5982\u4f55\u63d0\u5347\u751f\u6210\u5f0fAI\u6548\u80fd\u548c\u6548\u7387\uff1a<\/strong>Intel Gaudi 3 \u52a0\u901f\u5668\u5c08\u70ba\u9ad8\u6548\u7684\u5927\u898f\u6a21 AI \u904b\u7b97\u6253\u9020\uff0c\u63a1\u7528\u76f8\u8f03\u524d\u4e00\u4ee3\u7522\u54c1\u66f4\u5148\u9032\u7684 5 \u5948\u7c73\u88fd\u7a0b\u3002\u5176\u8a2d\u8a08\u5141\u8a31\u540c\u6642\u555f\u52d5\u6240\u6709\u5f15\u64ce\u4ee5\u63d0\u5347\u901f\u5ea6\uff0c\u5305\u62ec\u77e9\u9663\u4e58\u6cd5\u5f15\u64ce\uff08MME\uff09\u3001\u5f35\u91cf\u8655\u7406\u5668\u6838\u5fc3\uff08TPC\uff09 \u548c\u7db2\u8def\u4ecb\u9762\u5361\uff08NIC\uff09\uff0c\u9032\u800c\u5be6\u73fe\u66f4\u9ad8\u901f\u3001\u9ad8\u6548\u7684\u6df1\u5ea6\u5b78\u7fd2\u904b\u7b97\u548c\u898f\u6a21\u64f4\u5145\u3002Gaudi 3 \u52a0\u901f\u5668\u7684\u4e3b\u8981\u7279\u9ede\u5305\u62ec\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>AI\u5c08\u7528\u904b\u7b97\u5f15\u64ce\uff1a<\/strong>Intel Gaudi 3 \u52a0\u901f\u5668\u5c08\u70ba\u9ad8\u6548\u80fd\u3001\u9ad8\u6548\u7387\u7684\u751f\u6210\u5f0f AI \u904b\u7b97\u6240\u6253\u9020\u3002\u6bcf\u53f0\u52a0\u901f\u5668\u90fd\u6709\u5c08\u5c6c\u7684\u7570\u8cea\u904b\u7b97\u5f15\u64ce\uff0c\u7531 64 \u500b AI \u81ea\u8a02\u548c\u53ef\u7de8\u7a0b TPC \u548c 8 \u500b MME \u7d44\u6210\u3002\u6bcf\u500b Intel Gaudi 3 MME \u7686\u80fd\u57f7\u884c 64,000 \u500b\u5e73\u884c\u904b\u7b97\uff0c\u904b\u7b97\u6548\u7387\u6975\u9ad8\uff0c\u4e26\u64c5\u65bc\u8655\u7406\u8907\u96dc\u7684\u77e9\u9663\u904b\u7b97\uff0c\u9019\u4e5f\u662f\u6df1\u5ea6\u5b78\u7fd2\u6f14\u7b97\u6cd5\u7684\u57fa\u790e\u904b\u7b97\u3002\u6b64\u7368\u7279\u7684\u8a2d\u8a08\u5927\u5e45\u63d0\u5347\u5e73\u884c AI \u904b\u7b97\u7684\u901f\u5ea6\u548c\u6548\u7387\uff0c\u4e26\u652f\u63f4\u591a\u7a2e\u8cc7\u6599\u985e\u578b\uff0c\u5305\u62ec FP8 \u548c BF16\u3002<\/li>\n\n\n\n<li><strong>\u63d0\u5347\u8a18\u61b6\u9ad4\u5bb9\u91cf\uff0c\u6eff\u8db3LLM\u5bb9\u91cf\u9700\u6c42\uff1a<\/strong>Intel Gaudi 3 \u642d\u8f09 128 GB \u7684 HBMe2 \u8a18\u61b6\u9ad4\u5bb9\u91cf\u30013.7 TB \u7684\u8a18\u61b6\u9ad4\u983b\u5bec\u548c 96 MB \u7684 on-board \u975c\u614b\u96a8\u6a5f\u5b58\u53d6\u8a18\u61b6\u9ad4\uff08SRAM\uff09\uff0c\u56e0\u6b64\u80fd\u5920\u5728\u66f4\u5c11\u7684 Intel Gaudi 3 \u4e0a\uff0c\u63d0\u4f9b\u8655\u7406\u5927\u578b\u751f\u6210\u5f0f AI \u8cc7\u6599\u96c6\u6240\u9700\u7684\u8db3\u5920\u8a18\u61b6\u9ad4\uff0c\u4e14\u7279\u5225\u9069\u7528\u65bc\u5927\u578b\u8a9e\u8a00\u548c\u591a\u6a21\u614b\u6a21\u578b\uff0c\u6709\u52a9\u65bc\u63d0\u5347\u5de5\u4f5c\u8ca0\u8f09\u6548\u80fd\u548c\u8cc7\u6599\u4e2d\u5fc3\u7684\u6210\u672c\u6548\u7387\u3002<\/li>\n\n\n\n<li><strong>\u70ba\u4f01\u696d\u63d0\u4f9b\u751f\u6210\u5f0fAI\u9ad8\u6548\u7cfb\u7d71\u64f4\u5145\uff1a<\/strong>\u6bcf\u500b Intel Gaudi 3 \u52a0\u901f\u5668\u7686\u6574\u5408 24 \u500b 200 GB \u7684\u4e59\u592a\u7db2\u8def\u9023\u63a5\u57e0\uff0c\u63d0\u4f9b\u9748\u6d3b\u7684\u958b\u653e\u6a19\u6e96\u7db2\u8def\uff0c\u5be6\u73fe\u9ad8\u6548\u64f4\u5145\uff0c\u4ee5\u652f\u63f4\u5927\u578b\u904b\u7b97\u96c6\uff0c\u4e26\u514b\u670d\u5c08\u6709\u7db2\u8def\u67b6\u69cb\u7684\u4f9b\u61c9\u5546\u9650\u5236\u3002Intel Gaudi 3 \u52a0\u901f\u5668\u5be6\u73fe\u55ae\u4e00\u7bc0\u9ede\u5230\u4e0a\u5343\u7bc0\u9ede\u7684\u9ad8\u6548\u64f4\u5145\uff0c\u4ee5\u6eff\u8db3\u751f\u6210\u5f0f AI \u6a21\u578b\u7684\u5ee3\u6cdb\u8981\u6c42\u3002<\/li>\n\n\n\n<li><strong>\u958b\u653e\u7522\u696d\u8edf\u9ad4\u63d0\u5347\u958b\u767c\u4eba\u54e1\u751f\u7522\u529b\uff1a<\/strong>Intel Gaudi \u8edf\u9ad4\u6574\u5408 PyTorch \u6846\u67b6\uff0c\u4e26\u63d0\u4f9b\u57fa\u65bc Hugging Face \u793e\u7fa4\u7684\u6700\u4f73\u5316\u6a21\u578b\uff0c\u662f\u76ee\u524d\u751f\u6210\u5f0f AI \u958b\u767c\u4eba\u54e1\u6700\u5e38\u7528\u7684 AI \u6846\u67b6\uff0c\u4f7f\u751f\u6210\u5f0f AI \u958b\u767c\u4eba\u54e1\u80fd\u5920\u5728\u9ad8\u5ea6\u62bd\u8c61\u5c64\u4e0a\u9032\u884c\u64cd\u4f5c\uff0c\u63d0\u5347\u6613\u7528\u6027\u548c\u751f\u7522\u529b\uff0c\u4e26\u53ef\u8f15\u9b06\u5730\u5c07\u6a21\u578b\u8f49\u79fb\u5230\u4e0d\u540c\u786c\u9ad4\u985e\u578b\u4e0a\u3002<\/li>\n\n\n\n<li><strong>Gaudi 3 PCIe\uff1a<\/strong>Gaudi 3 \u9ad8\u901f PCIe \u9644\u52a0\u5361\u662f\u5168\u65b0\u7522\u54c1\uff0c\u5916\u578b\u898f\u683c\u5c08\u70ba\u5be6\u73fe\u9ad8\u6548\u7387\u4e26\u964d\u4f4e\u529f\u8017\u8a2d\u8a08\uff0c\u9069\u7528\u65bc\u5fae\u8abf\u3001\u63a8\u7406\u548c\u6aa2\u7d22\u589e\u5f37\u751f\u6210\uff08RAG\uff09\u7b49\u5de5\u4f5c\uff0c\u914d\u5099\u529f\u7387 600 \u74e6\u7684\u6a19\u6e96\uff08Full-height \uff09\u5c01\u88dd\uff0c128GB\u7684\u8a18\u61b6\u9ad4\u5bb9\u91cf\uff0c\u4e14\u983b\u5bec\u9054\u5230\u6bcf\u79d2 3.7TB\u3002<\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<p>Intel Gaudi 3 \u52a0\u901f\u5668\u5c07\u70ba\u9818\u5148\u751f\u6210\u5f0f AI \u6a21\u578b\u7684\u8a13\u7df4\u548c\u63a8\u7406\uff0c\u5e36\u4f86\u986f\u8457\u7684\u6548\u80fd\u63d0\u5347\u3002\u8207 Nvidia H100 \u76f8\u6bd4\uff0cGaudi 3 \u52a0\u901f\u5668\u7684\u5e73\u5747\u6548\u80fd\u9810\u671f\u5c07\u70ba\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u5728 Llama2 7B \u548c 13B \u53c3\u6578\u4ee5\u53caGPT-3 175B\u53c3\u6578\u6a21\u578b\u7684<strong>\u8a13\u7df4\u6642\u9593\u52a0\u5feb 50%<sup>1<\/sup>\u3002<\/strong><\/li>\n\n\n\n<li>\u5728 Llama 7B \u548c 70B \u4ee5\u53ca Falcon 180B \u53c3\u6578\u6a21\u578b\u4e0a\uff0c<strong>\u63a8\u8ad6\u541e\u5410\u91cf\u63d0\u5347 50%<sup>2<\/sup>\uff0c\u63a8\u8ad6\u80fd\u6e90\u6548\u7387\u63d0\u5347 40%<sup>3<\/sup><\/strong>\u3002\u5728\u8f03\u9577\u7684\u8f38\u5165\u548c\u8f38\u51fa\u5e8f\u5217\u4e2d\uff0c\u5177\u6709\u66f4\u5927\u7684\u63a8\u7406\u6548\u80fd\u512a\u52e2\u3002<\/li>\n\n\n\n<li>\u8207 Nvidia H200 \u76f8\u6bd4\uff0c\u5728 Llama 7B \u548c 70B \u4ee5\u53ca Falcon 180B \u53c3\u6578\u6a21\u578b\u7684<strong>\u63a8\u7406\u901f\u5ea6\u63d0\u5347 30%<sup>4<\/sup>\u3002<\/strong><\/li>\n<\/ul>\n\n\n\n<p><\/p>\n\n\n\n<p>Intel Gaudi 3 \u52a0\u901f\u5668\u5c07\u65bc 2024 \u5e74\u7b2c\u4e8c\u5b63\uff0c\u5411 OEM \u63d0\u4f9b\u901a\u7528\u57fa\u677f\u548c\u958b\u653e\u52a0\u901f\u5668\u6a21\u578b\uff08Open accelerator module, OAM\uff09\u7684\u696d\u754c\u6a19\u6e96\u914d\u7f6e\u3002\u5305\u542b\u6234\u723e\u79d1\u6280\u3001\u6167\u8207\u79d1\u6280\uff08HPE\uff09\u3001\u806f\u60f3\u548c\u7f8e\u8d85\u5fae\u7b49\u77e5\u540d OEM\uff0c\u90fd\u5c07\u63a1\u7528 Gaudi 3\u3002Intel Gaudi 3 \u52a0\u901f\u5668\u9810\u8a08\u65bc 2024 \u5e74\u7b2c\u4e09\u5b63\u5168\u9762\u4e0a\u5e02\uff0cIntel Gaudi 3 PCIe \u9644\u52a0\u5361\u9810\u8a08\u65bc 2024 \u5e74\u7b2c\u56db\u5b63\u4e0a\u5e02\u3002<\/p>\n\n\n\n<p>Intel Gaudi 3 \u52a0\u901f\u5668\u4e5f\u5c07\u652f\u63f4\u591a\u500b\u9ad8\u6210\u672c\u6548\u76ca LLM \u57fa\u790e\u67b6\u69cb\uff0c\u5354\u52a9\u8a13\u7df4\u548c\u63a8\u7406\uff0c\u4e26\u70ba\u5305\u62ec NAVER \u5728\u5167\u7684\u7d44\u7e54\u63d0\u4f9b\u6027\u50f9\u6bd4\u512a\u52e2\u548c\u9078\u64c7\u3002<\/p>\n\n\n\n<p>\u958b\u767c\u8005\u5f9e\u4eca\u65e5\u8d77\u5373\u53ef\u5b58\u53d6 Intel Developer Cloud \u4e0a<a href=\"https:\/\/developer.habana.ai\/intel-developer-cloud\/getting-started-on-the-intel-developer-cloud\/?utm_term=&amp;utm_campaign=PMax-+Google&amp;utm_source=adwords&amp;utm_medium=ppc&amp;hsa_acc=1034914560&amp;hsa_cam=21089989807&amp;hsa_grp=&amp;hsa_ad=&amp;hsa_src=x&amp;hsa_tgt=&amp;hsa_kw=&amp;hsa_mt=&amp;hsa_net=adwords&amp;hsa_ver=3&amp;gad_source=1&amp;gclid=Cj0KCQjw2PSvBhDjARIsAKc2cgNPd48lg6eFDhBo6dvmlXhZT20O25FioJwE17vlbrN6C86x51H4vhYaAl2LEALw_wcB\" target=\"_blank\" rel=\"noreferrer noopener\">\u4ee5Intel Gaudi 2\u70ba\u57fa\u790e\u7684\u5be6\u4f8b<\/a>\uff0c\u4ee5\u5b78\u7fd2\u3001\u5efa\u7acb\u539f\u578b\u3001\u6e2c\u8a66\u548c\u57f7\u884c\u61c9\u7528\u7a0b\u5f0f\u8207\u5de5\u4f5c\u8ca0\u8f09\u3002<\/p>\n\n\n\n<p>Intel Gaudi 3 \u52a0\u901f\u5668\u7684\u767c\u5c55\u5c07\u70ba\u82f1\u7279\u723e\u4e0b\u4e00\u4ee3\u91dd\u5c0d AI \u548c\u9ad8\u6548\u80fd\u904b\u7b97\u7684 GPU Falcon Shores \u5960\u4e0b\u57fa\u77f3\u3002Falcon Shores \u5c07\u6574\u5408 Intel Gaudi \u548c Intel\u00ae Xe \u7684\u667a\u6167\u8ca1\u7522\u6b0a\uff08IP\uff09\uff0c\u4ee5\u53ca\u5efa\u7acb\u5728 Intel\u00ae oneAPI \u898f\u7bc4\u7684\u55ae\u4e00 GPU \u53ef\u7a0b\u5f0f\u5316\u754c\u9762\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p>\u3000\u3000<\/p>\n\n\n\n<p><sup>1 NV H100 comparison based on: <a href=\"https:\/\/developer.nvidia.com\/deep-learning-performance-training-inference\/training\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/developer.nvidia.com\/deep-learning-performance-training-inference\/training<\/a>, Mar 28th 2024\u00a0 \u00e0 \u201cLarge Language Model\u201d tab Vs Intel\u00ae Gaudi\u00ae 3\u00a0 projections for LLAMA2-7B, LLAMA2-13B &amp; GPT3-175B as of 3\/28\/2024. Results may vary<br>2 NV H100 comparison based on <a href=\"https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html#h100-gpus-fp8<\/a> , Mar 28th, 2024. Reported numbers are per GPU. Vs Intel\u00ae Gaudi\u00ae 3 projections for LLAMA2-7B, LLAMA2-70B &amp; Falcon 180B projections. Results may vary.\u00a0<br>3 NV comparison based on <a href=\"https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html#h100-gpus-fp8<\/a> , Mar 28th, 2024. Reported numbers are per GPU. Vs Intel\u00ae Gaudi\u00ae 3 projections for LLAMA2-7B, LLAMA2-70B &amp; Falcon 180B Power efficiency for both Nvidia and Gaudi 3 based on internal estimates. Results may vary. \u00a0<br>4<strong> <\/strong>NV H200 comparison based on <a href=\"https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html\" target=\"_blank\" rel=\"noreferrer noopener\">https:\/\/nvidia.github.io\/TensorRT-LLM\/performance.html#h100-gpus-fp8<\/a> , Mar 28th, 2024. Reported numbers are per GPU.Vs Intel\u00ae Gaudi\u00ae 3 projections for LLAMA2-7B, LLAMA2-70B &amp; Falcon 180B projections. Results may vary.<\/sup><\/p>\n\n\n\n<p><\/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\">\ud83d\udfe6 \u770b\u7d55\u7f8eSeasonic Vertex GX-1000 SAKURA\u6afb\u82b1\u7248\u6e2c\u8a66\u6587 <a href=\"https:\/\/www.facebook.com\/ioioTIMES\/posts\/pfbid0GpUf8BxGtL6xmnv8s1gRJhw1GdYSyzwi5mPwRjWAA2coEp5bh53R4zNzcwU8TopLl\" target=\"_blank\" rel=\"noopener\">\u9001\u60a8\u96fb\u6e90\u4f9b\u61c9\u5668<\/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>\ud83d\udfe6<strong>\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","protected":false},"excerpt":{"rendered":"<p>\u5ef6\u7e8cGaudi 2\u7684\u6548\u80fd\u548c\u53ef\u64f4\u5145\u6027\uff0c<\/p>\n","protected":false},"author":3,"featured_media":64166,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[35,580,34,9252,9253,126],"class_list":["post-64155","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-ai","tag-focus","tag-intel","tag-intel-gaudi-3-ai","tag-vision-2024","tag-126"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/64155"}],"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=64155"}],"version-history":[{"count":5,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/64155\/revisions"}],"predecessor-version":[{"id":64224,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/64155\/revisions\/64224"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/64166"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=64155"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=64155"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=64155"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}