{"id":119561,"date":"2025-08-25T15:10:28","date_gmt":"2025-08-25T07:10:28","guid":{"rendered":"https:\/\/www.ioiotimes.com\/?p=119561"},"modified":"2025-08-25T15:11:49","modified_gmt":"2025-08-25T07:11:49","slug":"hot-chips%e5%a4%a7%e6%9c%83%e7%9a%84%e7%86%b1%e9%96%80%e8%a9%b1%e9%a1%8c%ef%bc%9a%e5%90%84%e7%a8%ae%e8%a6%8f%e6%a8%a1%e7%9a%84%e6%8e%a8%e8%ab%96%e3%80%81%e7%b6%b2%e8%b7%af%e8%88%87ai%e5%89%b5%e6%96%b0","status":"publish","type":"post","link":"https:\/\/www.ioiotimes.com\/?p=119561","title":{"rendered":"Hot Chips\u5927\u6703\u7684\u71b1\u9580\u8a71\u984c\uff1a\u5404\u7a2e\u898f\u6a21\u7684\u63a8\u8ad6\u3001\u7db2\u8def\u8207AI\u5275\u65b0\uff0c\u5168\u90fd\u5efa\u69cb\u65bcNVIDIA\u6280\u8853\u4e4b\u4e0a"},"content":{"rendered":"\n<h3 class=\"wp-block-heading\">\u5728\u7f8e\u570b\u52a0\u5dde\u5e15\u7f85\u5967\u5716\u8209\u884c\u7684Hot Chips\u5927\u6703\u4e0a\uff0cNVIDIA\u5c08\u5bb6\u5c07\u6df1\u5165\u89e3\u6790NVIDIA NVLink\u8207Spectrum-X\u4e59\u592a\u7db2\u8def\u6280\u8853\uff0c\u4ee5\u53caBlackwell\u8207CUDA\u5982\u4f55\u52a0\u901f\u5168\u7403\u6578\u767e\u842cAI\u5de5\u4f5c\u6d41\u7a0b\u7684\u63a8\u8ad6<\/h3>\n\n\n\n<p>\u4eba\u5de5\u667a\u6167\uff08AI\uff09\u63a8\u7406\u3001\u63a8\u8ad6\u53ca\u7db2\u8def\u6280\u8853\u5c07\u662f Hot Chips \u5927\u6703\u7684\u7126\u9ede\u8b70\u984c\u3002<\/p>\n\n\n\n<p>8 \u6708 24 \u81f3 26 \u65e5\u5728\u53f2\u4e39\u4f5b\u5927\u5b78\u8209\u8fa6\u7684 Hot Chips \u5927\u6703\uff0c\u662f\u7522\u696d\u8207\u5b78\u8853\u754c\u8655\u7406\u5668\u8207\u7cfb\u7d71\u67b6\u69cb\u4eba\u54e1\u7684\u91cd\u8981\u8ad6\u58c7\u6d3b\u52d5\uff0c\u5c55\u793a\u63a8\u52d5 <a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/ai-factory\/\" target=\"_blank\" rel=\"noreferrer noopener\">AI \u5de5\u5ee0<\/a>\u767c\u5c55\u53ca\u9a45\u52d5\u4e0a\u5146\u7f8e\u5143\u7b49\u7d1a\u8cc7\u6599\u4e2d\u5fc3\u904b\u7b97\u5e02\u5834\u7684\u6700\u65b0\u5275\u65b0\u6280\u8853\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1200\" height=\"675\" src=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2025\/08\/20250825-nvidia02.jpg\" alt=\"\" class=\"wp-image-119562\" title=\"\" srcset=\"https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2025\/08\/20250825-nvidia02.jpg 1200w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2025\/08\/20250825-nvidia02-300x169.jpg 300w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2025\/08\/20250825-nvidia02-1024x576.jpg 1024w, https:\/\/www.ioiotimes.com\/wordpress\/wp-content\/uploads\/2025\/08\/20250825-nvidia02-768x432.jpg 768w\" sizes=\"(max-width: 1200px) 100vw, 1200px\" \/><\/figure>\n\n\n\n<p><\/p>\n\n\n\n<p>\u5728\u9019\u5834\u76db\u6703\u4e0a\uff0cNVIDIA \u8207 Google\u3001\u5fae\u8edf\u7b49\u7522\u696d\u9818\u8896\u4e00\u540c\u53c3\u8207 8 \u6708 24 \u65e5\u8209\u884c\u7684\u300c\u6559\u5b78\uff08tutorial\uff09\u300d\u6d3b\u52d5\uff0c\u63a2\u8a0e\u8cc7\u6599\u4e2d\u5fc3\u7684\u6a5f\u67b6\u7d1a\u67b6\u69cb\u8a2d\u8a08\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0cNVIDIA \u5c08\u5bb6\u5718\u968a\u9084\u5c07\u5728\u56db\u5834\u6f14\u8b1b\u53ca\u4e00\u5834\u6559\u5b78\u6d3b\u52d5\u8a73\u7d30\u8aaa\u660e\uff1a<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u3000\u25cf\u3000\u5305\u62ec <a href=\"https:\/\/www.nvidia.com\/zh-tw\/networking\/products\/ethernet\/supernic\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA ConnectX-8 SuperNIC<\/a> \u5728\u5167\u7684 NVIDIA \u7db2\u8def\u6280\u8853\uff0c\u5982\u4f55\u5728\u6a5f\u67b6\u8207\u8cc7\u6599\u4e2d\u5fc3\u898f\u6a21\u4e0b\u5be6\u73fe AI \u63a8\u7406\uff08\u7531 NVIDIA \u7db2\u8def\u4ecb\u9762\u5361\u8207\u7cfb\u7d71\u55ae\u6676\u7247\u9996\u5e2d\u67b6\u69cb\u5e2b Idan Burstein \u4e3b\u8b1b\uff09\u3002<\/li>\n\n\n\n<li>\u3000\u25cf\u3000\u7531\u5305\u62ec <a href=\"https:\/\/www.nvidia.com\/zh-tw\/geforce\/graphics-cards\/50-series\/rtx-5090\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA GeForce RTX 5090 GPU<\/a> \u5728\u5167\u7684 NVIDIA Blackwell \u67b6\u69cb\u6240\u5e36\u52d5\u7684\u795e\u7d93\u6e32\u67d3\u6280\u8853\u9032\u6b65\u8207\u63a8\u7406\u6548\u80fd\u5927\u5e45\u8e8d\u9032\uff0c\u5c07\u7e6a\u5716\u8207\u6a21\u64ec\u80fd\u529b\u63d0\u5347\u5230\u5168\u65b0\u9ad8\u5ea6\uff08\u7531 NVIDIA \u67b6\u69cb\u8cc7\u6df1\u7e3d\u76e3 Marc Blackstein \u4e3b\u8b1b\uff09\u3002<\/li>\n\n\n\n<li>\u3000\u25cf\u3000\u6574\u5408\u77fd\u5149\u5b50\u6280\u8853\u7684<a href=\"https:\/\/www.nvidia.com\/en-us\/networking\/products\/silicon-photonics\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u5171\u540c\u5c01\u88dd\u5149\u5b78\uff08CPO\uff09\u4ea4\u63db\u5668<\/a>\u4ee5\u5149\u7e96\u53d6\u4ee3\u50b3\u7d71\u9285\u7dda\uff0c\u8cc7\u6599\u50b3\u8f38\u7684\u901f\u5ea6\u66f4\u5feb\u3001\u80fd\u6e90\u6d88\u8017\u4e5f\u66f4\u5c11\uff0c\u53ef\u5be6\u73fe\u5efa\u7f6e\u9ad8\u6548\u7387\u3001\u9ad8\u6548\u80fd\u3001\u5343\u5146\u74e6\u898f\u6a21\u7684 AI \u5de5\u5ee0\u3002\u672c\u5834\u6f14\u8b1b\u4e5f\u5c07\u4ecb\u7d39 <a href=\"https:\/\/nvidianews.nvidia.com\/news\/nvidia-introduces-spectrum-xgs-ethernet-to-connect-distributed-data-centers-into-giga-scale-ai-super-factories\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Spectrum-XGS \u4e59\u592a\u7db2\u8def\u6280\u8853<\/a>\uff0c\u9019\u662f\u4e00\u9805\u5168\u65b0\u7684\u8de8\u57df\u64f4\u5c55\u6280\u8853\uff0c\u53ef\u5c07\u5206\u6563\u5f0f\u8cc7\u6599\u4e2d\u5fc3\u7d71\u4e00\u70ba AI \u8d85\u7d1a\u5de5\u5ee0\uff08\u7531 NVIDIA \u7db2\u8def\u90e8\u9580\u8cc7\u6df1\u526f\u7e3d\u88c1 Gilad Shainer \u4e3b\u8b1b\uff09\u3002<\/li>\n\n\n\n<li>\u3000\u25cf\u3000NVIDIA GB10 \u8d85\u7d1a\u6676\u7247\u662f <a href=\"https:\/\/www.nvidia.com\/zh-tw\/products\/workstations\/dgx-spark\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA DGX Spark<\/a> \u684c\u4e0a\u578b\u8d85\u7d1a\u96fb\u8166\u7684\u5f15\u64ce\uff08\u7531 NVIDIA \u8cc7\u6df1\u5091\u51fa\u5de5\u7a0b\u5e2b Andi Skende \u4e3b\u8b1b\uff09\u3002<\/li>\n<\/ul>\n\n\n\n<p>\u9019\u4e00\u5207\u90fd\u9ad4\u73fe\u4e86 NVIDIA \u7684\u6700\u65b0\u6280\u8853\u5982\u4f55\u52a0\u901f\u63a8\u8ad6\uff0c\u5f9e\u800c\u63a8\u52d5\u7121\u6240\u4e0d\u5728\u3001\u6db5\u84cb\u5404\u7a2e\u898f\u6a21\u7684 AI \u5275\u65b0\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>NVIDIA\u7db2\u8def\u6280\u8853\u63a8\u52d5\u5927\u898f\u6a21AI\u5275\u65b0<\/strong><\/h3>\n\n\n\n<p><a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/ai-reasoning\/\" target=\"_blank\" rel=\"noreferrer noopener\">AI \u63a8\u7406<\/a>\uff0c\u4e5f\u5c31\u662f AI \u7cfb\u7d71\u900f\u904e\u591a\u91cd AI \u63a8\u7406\u6d41\u7a0b\u5206\u6790\u4e26\u89e3\u6c7a\u8907\u96dc\u554f\u984c\uff0c\u9700\u8981\u6a5f\u67b6\u7b49\u7d1a\u7684\u904b\u7b97\u6548\u80fd\uff0c\u624d\u80fd\u9ad8\u6548\u7387\u5730\u63d0\u4f9b\u6700\u4f73\u7684\u4f7f\u7528\u8005\u9ad4\u9a57\u3002<\/p>\n\n\n\n<p>\u5728\u652f\u63f4\u7576\u524d AI \u5de5\u4f5c\u8ca0\u8f09\u7684\u8cc7\u6599\u4e2d\u5fc3\u4e2d\uff0c\u7db2\u8def\u626e\u6f14\u8457\u4e2d\u6a1e\u795e\u7d93\u7cfb\u7d71\u7684\u89d2\u8272\uff0c\u5c07\u4f3a\u670d\u5668\u3001\u5132\u5b58\u88dd\u7f6e\u8207\u5176\u4ed6\u786c\u9ad4\u7b49\u6240\u6709\u5143\u4ef6\uff0c\u4e32\u9023\u6210\u70ba\u4e00\u500b\u5b8c\u6574\u3001\u7dca\u5bc6\u7d50\u5408\u53c8\u5f37\u5927\u7684\u904b\u7b97\u55ae\u5143\u3002<\/p>\n\n\n\n<p>Burstein \u5728 Hot Chips \u5927\u6703\u7684\u6f14\u8b1b\u6d3b\u52d5\u5c07\u6df1\u5165\u63a2\u8a0e NVIDIA \u7684\u7db2\u8def\u6280\u8853\uff0c\u7279\u5225\u662f NVIDIA ConnectX-8 SuperNIC\uff0c\u5982\u4f55\u5be6\u73fe\u9ad8\u901f\u3001\u4f4e\u5ef6\u9072\u7684\u591a GPU \u901a\u8a0a\uff0c\u4ee5\u63d0\u4f9b\u9818\u5148\u696d\u754c\u7684\u5927\u898f\u6a21 AI \u63a8\u7406\u6548\u80fd\u3002<\/p>\n\n\n\n<p>\u4f5c\u70ba NVIDIA \u7db2\u8def\u5e73\u53f0\u7684\u4e00\u90e8\u5206\uff0cNVIDIA NVLink\u3001NVLink Switch \u8207 NVLink Fusion \u63d0\u4f9b\u5782\u76f4\u64f4\u5c55\u7684\u9023\u7d50\u80fd\u529b\uff0c\u9023\u63a5\u4f3a\u670d\u5668\u5167\u90e8\u53ca\u4f3a\u670d\u5668\u9593\u7684 GPU \u8207\u904b\u7b97\u5143\u4ef6\uff0c\u5be6\u73fe\u8d85\u4f4e\u5ef6\u9072\u8207\u9ad8\u983b\u5bec\u7684\u8cc7\u6599\u4ea4\u63db\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/www.nvidia.com\/zh-tw\/networking\/spectrumx\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Spectrum-X \u4e59\u592a\u7db2\u8def<\/a> \u5247\u63d0\u4f9b\u6c34\u5e73\u64f4\u5c55\u7684\u67b6\u69cb\uff0c\u80fd\u5c07\u6574\u500b\u53e2\u96c6\u9023\u7d50\u8d77\u4f86\uff0c\u5feb\u901f\u5c07\u5927\u91cf\u8cc7\u6599\u96c6\u4e32\u6d41\u81f3 AI \u6a21\u578b\uff0c\u4e26\u5728\u8cc7\u6599\u4e2d\u5fc3\u5167\u5354\u8abf GPU \u8207 GPU \u4e4b\u9593\u7684\u901a\u8a0a\u3002<a href=\"https:\/\/nvidianews.nvidia.com\/news\/nvidia-introduces-spectrum-xgs-ethernet-to-connect-distributed-data-centers-into-giga-scale-ai-super-factories\" target=\"_blank\" rel=\"noreferrer noopener\">Spectrum-XGS \u4e59\u592a\u7db2\u8def<\/a>\u7684\u8de8\u57df\u64f4\u5c55\u6280\u8853\uff0c\u66f4\u9032\u4e00\u6b65\u64f4\u5c55\u4e86 Spectrum-X \u7684\u6975\u81f4\u6548\u80fd\u8207\u898f\u6a21\uff0c\u5c07\u591a\u500b\u5206\u6563\u5f0f\u8cc7\u6599\u4e2d\u5fc3\u9032\u884c\u4e92\u9023\uff0c\u5f9e\u800c\u6253\u9020\u51fa\u5177\u5099\u5343\u5146\u7d1a\u667a\u6167\u80fd\u529b\u7684 AI \u8d85\u7d1a\u5de5\u5ee0\u3002<\/p>\n\n\n\n<p>\u4f5c\u70ba Spectrum-X \u4e59\u592a\u7db2\u8def\u7684\u6838\u5fc3\uff0cCPO \u4ea4\u63db\u5668\u7a81\u7834\u5927\u898f\u6a21 AI \u57fa\u790e\u8a2d\u65bd\u7684\u6548\u80fd\u8207\u6548\u7387\u6975\u9650\uff0cShainer \u5c07\u5728\u6f14\u8b1b\u4e2d\u91dd\u5c0d\u9019\u9805\u6280\u8853\u8a73\u7d30\u4ecb\u7d39\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/www.nvidia.com\/zh-tw\/data-center\/gb200-nvl72\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA GB200 NVL72<\/a> \u662f\u63a1\u7528\u55ae\u4e00\u6a5f\u67b6\u7684\u767e\u842c\u5146\u7d1a\uff08exascale\uff09\u96fb\u8166\uff0c\u914d\u5099 36 \u500b NVIDIA GB200 \u8d85\u7d1a\u6676\u7247\uff0c\u6bcf\u500b\u6676\u7247\u6574\u5408\u5169\u500b NVIDIA B200 GPU \u8207\u4e00\u500b NVIDIA Grace CPU\uff0c\u9019\u4e9b\u904b\u7b97\u55ae\u5143\u900f\u904e\u6709\u53f2\u4ee5\u4f86\u6700\u5927\u7684 NVLink \u7db2\u57df\u5f7c\u6b64\u4e92\u9023\uff0c\u4e26\u7531 NVLink Switch \u70ba AI \u8207\u9ad8\u6548\u80fd\u904b\u7b97\u5de5\u4f5c\u8ca0\u8f09\u63d0\u4f9b\u6bcf\u79d2 130 TB \u7684\u4f4e\u5ef6\u9072 GPU \u901a\u8a0a\u3002<\/p>\n\n\n\n<p>\u57fa\u65bc NVIDIA Blackwell \u67b6\u69cb\u6253\u9020\u7684 GB200 NVL72 \u7cfb\u7d71\uff0c\u5728\u63a8\u7406\u8207\u63a8\u8ad6\u6548\u80fd\u65b9\u9762\u5be6\u73fe\u5927\u5e45\u8e8d\u9032\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>NVIDIA Blackwell\u8207CUDA\u5c07AI\u5e36\u7d66\u6578\u767e\u842c\u958b\u767c\u8005<\/strong><\/h3>\n\n\n\n<p>Blackstein \u5c07\u5728\u6f14\u8b1b\u4e2d\u6df1\u5165\u4ecb\u7d39\u540c\u6a23\u63a1\u7528 Blackwell \u67b6\u69cb\u7684 NVIDIA GeForce RTX 5090 GPU\uff0c\u900f\u904e <a href=\"https:\/\/www.nvidia.com\/zh-tw\/geforce\/technologies\/dlss\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA DLSS 4<\/a> \u6280\u8853\uff0c\u4f7f\u7576\u4eca\u904a\u6232\u7684\u6548\u80fd\u7ffb\u500d\u3002<\/p>\n\n\n\n<p>NVIDIA DLSS 4 \u6280\u8853\u9084\u80fd\u70ba\u904a\u6232\u52a0\u5165\u795e\u7d93\u6e32\u67d3\u529f\u80fd\uff0c\u5f9e\u800c\u5c07\u6548\u80fd\u63d0\u5347\u9ad8\u9054 10 \u500d\u3001\u8ffd\u8de1\u6548\u679c\u653e\u5927 10 \u500d\u53ca\u8a2d\u8a08\u9031\u671f\u7e2e\u77ed 10 \u500d\uff0c\u9032\u4e00\u6b65\u5f37\u5316\u96fb\u8166\u5716\u5f62\u8207\u6a21\u64ec\u7684\u771f\u5be6\u611f\u3002\u9019\u4e0d\u50c5\u80fd\u4ee5\u66f4\u4f4e\u80fd\u8017\u63d0\u4f9b\u6d41\u66a2\u800c\u9748\u654f\u7684\u8996\u89ba\u9ad4\u9a57\uff0c\u89d2\u8272\u8207\u7279\u6548\u7684\u903c\u771f\u7a0b\u5ea6\u4e5f\u53ef\u5927\u5e45\u63d0\u5347\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/developer.nvidia.com\/cuda-toolkit\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA CUDA<\/a> \u662f\u5168\u7403\u61c9\u7528\u6700\u666e\u53ca\u7684\u904b\u7b97\u57fa\u790e\u67b6\u69cb\uff0c\u8b93\u4f7f\u7528\u8005\u80fd\u5920\u5728\u4efb\u4f55\u5730\u65b9\u4f7f\u7528 NVIDIA Blackwell \u90e8\u7f72\u4e26\u904b\u884c AI \u6a21\u578b\u3002<\/p>\n\n\n\n<p>\u5168\u7403\u5df2\u6709\u6578\u5104\u7684 GPU \u904b\u884c CUDA\uff0c\u5f9e NVIDIA GB200 NVL72 \u6a5f\u67b6\u898f\u6a21\u7cfb\u7d71\uff0c\u4e00\u76f4\u5230\u642d\u8f09 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/geforce\/graphics-cards\/50-series\/rtx-5090\/\" target=\"_blank\" rel=\"noreferrer noopener\">GeForce RTX<\/a> \u53ca <a href=\"https:\/\/www.nvidia.com\/zh-tw\/products\/workstations\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA RTX PRO<\/a> \u7684\u500b\u4eba\u96fb\u8166\u8207\u5de5\u4f5c\u7ad9\uff0c\u800c\u5728 Skende \u7684\u6f14\u8b1b\u4e2d\u8a0e\u8ad6\u3001NVIDIA GB10 \u9a45\u52d5\u7684 <a href=\"https:\/\/www.nvidia.com\/zh-tw\/products\/workstations\/dgx-spark\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA DGX Spark<\/a> \u4e5f\u5373\u5c07\u63a8\u51fa\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u5f9e\u6f14\u7b97\u6cd5\u5230AI\u8d85\u7d1a\u96fb\u8166\uff0c\u70ba\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u6700\u4f73\u5316\u800c\u751f<\/strong><\/h3>\n\n\n\n<p>DGX Spark \u4ee5\u7cbe\u5de7\u7684\u5916\u578b\uff0c\u63d0\u4f9b\u5f37\u5927\u7684\u6548\u80fd\u8207\u529f\u80fd\uff0c\u8b93\u958b\u767c\u4eba\u54e1\u3001\u7814\u7a76\u4eba\u54e1\u3001\u8cc7\u6599\u79d1\u5b78\u5bb6\u8207\u5b78\u751f\u80fd\u5728\u684c\u9762\u7aef\u7a81\u7834<a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/generative-ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u751f\u6210\u5f0f AI<\/a> \u7684\u6975\u9650\uff0c\u4e26\u4e14\u52a0\u901f\u63a8\u52d5\u5404\u7522\u696d\u8655\u7406\u5de5\u4f5c\u8ca0\u8f09\u3002<\/p>\n\n\n\n<p>\u4f5c\u70ba NVIDIA Blackwell \u5e73\u53f0\u7684\u4e00\u90e8\u5206\uff0cDGX Spark \u652f\u63f4 NVFP4\uff0cNVFP4 \u662f\u4e00\u7a2e\u4f4e\u7cbe\u5ea6\u6578\u503c\u683c\u5f0f\uff0c\u53ef\u5be6\u73fe\u9ad8\u6548\u7684<a href=\"https:\/\/blogs.nvidia.com\/blog\/what-is-agentic-ai\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u4ee3\u7406\u578b AI<\/a> \u63a8\u8ad6\uff0c\u7279\u5225\u662f<a href=\"https:\/\/www.nvidia.com\/en-us\/glossary\/large-language-models\/\" target=\"_blank\" rel=\"noreferrer noopener\">\u5927\u578b\u8a9e\u8a00\u6a21\u578b\uff08LLM\uff09<\/a>\u3002\u5982\u6b32\u4e86\u89e3\u66f4\u591a\u95dc\u65bc NVFP4 \u7684\u8cc7\u8a0a\uff0c\u8acb\u53c3\u95b1 NVIDIA \u6280\u8853\u90e8\u843d\u683c\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<h3 class=\"wp-block-heading\"><strong>\u958b\u6e90\u5354\u4f5c\u63a8\u52d5\u63a8\u8ad6\u5275\u65b0<\/strong><strong><\/strong><\/h3>\n\n\n\n<p>NVIDIA \u52a0\u901f\u4e86\u591a\u9805\u958b\u6e90\u51fd\u5f0f\u5eab\u8207\u6846\u67b6\uff0c\u4ee5\u52a0\u901f\u4e26\u6700\u4f73\u5316\u5927\u578b\u8a9e\u8a00\u6a21\u578b\u8207\u5206\u6563\u5f0f\u63a8\u8ad6\u7684 AI \u5de5\u4f5c\u8ca0\u8f09\u3002\u9019\u4e9b\u51fd\u5f0f\u5eab\u8207\u6846\u67b6\u5305\u62ec <a href=\"https:\/\/docs.nvidia.com\/tensorrt-llm\/index.html\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA TensorRT-LLM<\/a>\u3001<a href=\"https:\/\/www.nvidia.com\/en-us\/ai\/dynamo\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Dynamo<\/a>\u3001TileIR\u3001Cutlass\u3001<a href=\"https:\/\/developer.nvidia.com\/nccl\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA Collective Communication Library<\/a> \u53ca NIX\uff0c\u76ee\u524d\u5df2\u6574\u5408\u81f3\u6578\u767e\u842c\u500b\u5de5\u4f5c\u6d41\u7a0b\u4e2d\u3002<\/p>\n\n\n\n<p>\u70ba\u4e86\u8b93\u958b\u767c\u8005\u80fd\u4ee5\u81ea\u884c\u9078\u64c7\u7684\u6846\u67b6\u9032\u884c\u5efa\u7f6e\uff0cNVIDIA \u8207\u9802\u5c16\u7684\u958b\u6e90\u6846\u67b6\u4f9b\u61c9\u5546\u5408\u4f5c\uff0c\u70ba FlashInfer\u3001PyTorch\u3001SGLang\u3001vLLM \u7b49\u63d0\u4f9b\u6a21\u578b\u6700\u4f73\u5316\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c<a href=\"https:\/\/www.nvidia.com\/zh-tw\/ai-data-science\/products\/nim-microservices\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA NIM \u5fae\u670d\u52d9<\/a>\u4e5f\u5df2\u652f\u63f4 OpenAI \u7684 gpt-oss\u3001Llama 4 \u7b49\u591a\u500b\u71b1\u9580\u7684\u958b\u6e90\u6a21\u578b\uff0c\u8b93\u958b\u767c\u8005\u80fd\u8f15\u9b06\u64cd\u4f5c\u8a17\u7ba1\u578b API\uff0c\u4e26\u5728\u5176\u9996\u9078\u7684\u57fa\u790e\u67b6\u69cb\u4e0a\u4eab\u6709\u81ea\u6211\u8a17\u7ba1\u6a21\u578b\u7684\u9748\u6d3b\u6027\u8207\u5b89\u5168\u6027\u3002<\/p>\n\n\n\n<p>\u6b61\u8fce\u53c3\u52a0 <a href=\"https:\/\/hotchips.org\/\" target=\"_blank\" rel=\"noreferrer noopener\">NVIDIA \u5728 Hot Chips \u5927\u6703\u7684\u5404\u9805\u6d3b\u52d5<\/a>\uff0c\u4ee5\u4e86\u89e3\u66f4\u591a\u95dc\u65bc\u63a8\u8ad6\u8207\u52a0\u901f\u904b\u7b97\u9818\u57df\u7684\u6700\u65b0\u9032\u5c55\u3002<\/p>\n\n\n\n<p><\/p>\n\n\n\n<p>\u3000\u3000\u3000\u3000<\/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>\ud83d\udfe6\u73fe\u5728\u5c31\u52a0\u5165&nbsp;ioioTIMES \u81c9\u66f8\u7c89\u7d72\u5718&nbsp;\u66f4\u591a\u4e92\u52d5\u3001\u66f4\u591a\u597d\u5eb7\u650f\u62b5\u52a0!!<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<\/strong>\u624b<br><\/h4>\n","protected":false},"excerpt":{"rendered":"<p>\u5728\u7f8e\u570b\u52a0\u5dde\u5e15\u7f85\u5967\u5716\u8209\u884c\u7684Hot Ch<\/p>\n","protected":false},"author":3,"featured_media":119562,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[13],"tags":[580,10748,71,14730,11586],"class_list":["post-119561","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news","tag-focus","tag-hot-chips","tag-nvidia","tag-nvlink","tag-spectrum-x"],"_links":{"self":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/119561"}],"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=119561"}],"version-history":[{"count":2,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/119561\/revisions"}],"predecessor-version":[{"id":119564,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/posts\/119561\/revisions\/119564"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=\/wp\/v2\/media\/119562"}],"wp:attachment":[{"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=119561"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=119561"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.ioiotimes.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=119561"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}