{"id":710,"date":"2025-01-29T12:02:22","date_gmt":"2025-01-29T12:02:22","guid":{"rendered":"https:\/\/janusai.pro\/?p=710"},"modified":"2025-01-29T12:02:40","modified_gmt":"2025-01-29T12:02:40","slug":"deepseek-v3-paper-details-how-to-bypass-the-cuda-monopoly","status":"publish","type":"post","link":"https:\/\/janusai.pro\/zh\/deepseek-v3-paper-details-how-to-bypass-the-cuda-monopoly\/","title":{"rendered":"DeepSeek V3 \u8ad6\u6587\u8a73\u7d30\u5167\u5bb9\uff1a\u5982\u4f55\u7e5e\u904e CUDA \u58df\u65b7\uff01"},"content":{"rendered":"<div style=\"margin-top: 0px; margin-bottom: 0px;\" class=\"sharethis-inline-share-buttons\" ><\/div>\n<p><a href=\"https:\/\/www.deepseek.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">DeepSeek V3<\/a> \u8ad6\u6587\u8a73\u60c5\uff1a\u5982\u4f55\u7e5e\u904e CUDA \u7684\u58df\u65b7\uff01<\/p>\n\n\n\n<p>DeepSeek \u6700\u8fd1\u767c\u5e03\u7684\u5169\u6b3e\u6a5f\u578b DeepSeek-V3 \u548c DeepSeek-R1\uff0c\u4ee5\u66f4\u4f4e\u7684\u6210\u672c\u5be6\u73fe\u4e86\u8207 OpenAI \u985e\u4f3c\u6a5f\u578b\u76f8\u5ab2\u7f8e\u7684\u6027\u80fd\u3002<\/p>\n\n\n\n<p>\u6839\u64da\u5916\u570b\u5a92\u9ad4\u5831\u5c0e\uff0c\u5728\u77ed\u77ed\u5169\u500b\u6708\u5167\uff0c\u4ed6\u5011\u5728\u7531 2,048 \u500b H800 GPU \u7d44\u6210\u7684\u96c6\u7fa4\u4e0a\uff0c\u8a13\u7df4\u51fa\u4e86\u4e00\u500b\u64c1\u6709 6,710 \u5104\u500b\u53c3\u6578\u7684 MoE \u8a9e\u8a00\u6a21\u578b\uff0c\u6548\u7387\u662f\u9802\u5c16 AI \u7684 10 \u500d\u3002<\/p>\n\n\n\n<p>\u9019\u9805\u7a81\u7834\u4e26\u975e\u4f7f\u7528 CUDA \u5be6\u73fe\uff0c\u800c\u662f\u900f\u904e\u5927\u91cf\u7684\u7d30\u7c92\u5ea6\u6700\u4f73\u5316\uff0c\u4ee5\u53ca\u4f7f\u7528 NVIDIA \u985e\u4f3c\u7d44\u5408\u7684 PTX (\u5e73\u884c\u7dda\u7a0b\u57f7\u884c) \u7a0b\u5f0f\u8a2d\u8a08\u3002<\/p>\n\n\n\n<p><a href=\"https:\/\/www.deepseek.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">\u6df1\u5ea6\u641c\u5c0b<\/a> \u88ab\u8feb\u8d70\u4e0a\u4e86\u4e00\u689d\u8207 OpenAI \u548c\u5176\u4ed6\u516c\u53f8\u4e0d\u540c\u7684\u9053\u8def\uff0c\u9019\u4e9b\u516c\u53f8\u5728\u786c\u9ad4\u9650\u5236\u4e0b\u4f9d\u8cf4\u7c97\u66b4\u7684\u904b\u7b97\u80fd\u529b\u3002\u5b83\u904b\u7528\u4e86\u4e00\u7cfb\u5217\u7684\u6280\u8853\u5275\u65b0\u4f86\u964d\u4f4e\u6a21\u578b\u7684\u529f\u8017\u9700\u6c42\uff0c\u540c\u6642\u9054\u5230\u6548\u80fd\u7684\u63d0\u5347\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=ZDM1YTM0ODZkYmQzOWNkNzc2ZTBmNzUwY2ZjOWYxMjZfYnUyVHFsb05ya0c1M0hvMGRUbk9CN3FVekR1ZjlQMEZfVG9rZW46TUtzM2JudThpb1p3NHJ4SlZNeWNWdU10bnNnXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u4e00\u4e9b\u7db2\u53cb\u7684\u71b1\u70c8\u8a55\u8ad6\uff1a<\/p>\n\n\n\n<p>\"\u5728\u9019\u500b\u4e16\u754c\u4e0a\uff0c\u5982\u679c\u6709\u4efb\u4f55\u4e00\u7fa4\u4eba\u6703\u760b\u72c2\u5730\u8aaa\u300cCUDA \u592a\u6162\u4e86\u300d\u9019\u6a23\u7684\u8a71\uff01<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=OGEyMmE3ZTJkODlkZDlmNjliZTI1MzI5YTE4ZWE3MjdfWWRBam5VTkVaV1ZsMFg3VzVTRjRDZlUzV2ZiSHZYT2RfVG9rZW46VGZsdWJrTzZHb243OUx4bEZsbmNmMFNzblFiXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=NzI1ZTBlMjJkMDI2N2MyMDdkMGI4YmU5OTJjNGM0YzFfZW4xbjVERFdhdGVObHBDUWR3NVZjbDRSM2lrVDlWRGlfVG9rZW46Q2N5MWIxV2ltbzdmZU14VXI2amNuZDk2bmRkXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=MDMyN2YyYjYwYTNkZDhmMmEyYWY2MjMzZGE3MGM1ZmFfM29veUZrRWdYODRGR0JVdWVVTnRoMzVwTWxjV09CT25fVG9rZW46SVE2dGJWek9Mb29jaTJ4ZnkzWWN5bUZWbnVnXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_82_2 counter-hierarchy ez-toc-counter ez-toc-grey ez-toc-container-direction\">\n<div class=\"ez-toc-title-container\">\n<p class=\"ez-toc-title\" style=\"cursor:inherit\">\u76ee\u9304<\/p>\n<span class=\"ez-toc-title-toggle\"><a href=\"#\" class=\"ez-toc-pull-right ez-toc-btn ez-toc-btn-xs ez-toc-btn-default ez-toc-toggle\" aria-label=\"\u5207\u63db\u5167\u5bb9\u8868\"><span class=\"ez-toc-js-icon-con\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">\u5207\u63db<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #999;color:#999\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewbox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #999;color:#999\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewbox=\"0 0 24 24\" version=\"1.2\" baseprofile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/span><\/a><\/span><\/div>\n<nav><ul class='ez-toc-list ez-toc-list-level-1' ><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/janusai.pro\/zh\/deepseek-v3-paper-details-how-to-bypass-the-cuda-monopoly\/#Genius_geeks_fine-tune_PTX_to_maximize_GPU_performance\" >\u5929\u624d\u6975\u5ba2\u5fae\u8abf PTX\uff0c\u5c07 GPU \u6548\u80fd\u767c\u63ee\u5230\u6975\u81f4<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/janusai.pro\/zh\/deepseek-v3-paper-details-how-to-bypass-the-cuda-monopoly\/#PTX_and_CUDA\" >PTX \u548c CUDA<\/a><\/li><li class='ez-toc-page-1 ez-toc-heading-level-2'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/janusai.pro\/zh\/deepseek-v3-paper-details-how-to-bypass-the-cuda-monopoly\/#However_the_technical_barriers_remain\" >\u7136\u800c\uff0c\u6280\u8853\u969c\u7919\u4ecd\u7136\u5b58\u5728<\/a><\/li><\/ul><\/nav><\/div>\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"Genius_geeks_fine-tune_PTX_to_maximize_GPU_performance\"><\/span>\u5929\u624d\u6975\u5ba2\u5fae\u8abf PTX\uff0c\u5c07 GPU \u6548\u80fd\u767c\u63ee\u5230\u6975\u81f4<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>NVIDIA PTX (Parallel Thread Execution\uff0c\u5e73\u884c\u7dda\u7a0b\u57f7\u884c) \u662f\u5c08\u70ba GPU \u8a2d\u8a08\u7684\u4e2d\u9593\u6307\u4ee4\u96c6\u67b6\u69cb\uff0c\u4ecb\u65bc\u9ad8\u968e GPU \u7a0b\u5f0f\u8a9e\u8a00 (\u4f8b\u5982 CUDA C\/C++) \u6216\u5176\u4ed6\u8a9e\u8a00\u524d\u7aef\u8207\u4f4e\u968e\u6a5f\u5668\u7a0b\u5f0f\u78bc (streaming assembly \u6216 SASS) \u4e4b\u9593\u3002<\/p>\n\n\n\n<p>PTX \u662f\u4e00\u7a2e\u4f4e\u968e\u6307\u4ee4\u96c6\u67b6\u69cb\uff0c\u5c07 GPU \u5448\u73fe\u70ba\u8cc7\u6599\u5e73\u884c\u904b\u7b97\u88dd\u7f6e\uff0c\u53ef\u9032\u884c\u5bc4\u5b58\u5668\u5206\u914d\u548c\u7dda\u7a0b\/\u7dda\u7a0b\u53e2\u96c6\u5c64\u7d1a\u8abf\u6574\u7b49\u7d30\u7c92\u5ea6\u6700\u4f73\u5316\uff0c\u9019\u4e9b\u90fd\u662f CUDA C\/C++ \u7b49\u8a9e\u8a00\u7121\u6cd5\u5be6\u73fe\u7684\u3002<\/p>\n\n\n\n<p>\u7576 PTX \u8f49\u63db\u70ba SASS \u6642\uff0c\u6703\u91dd\u5c0d\u7279\u5b9a\u4e16\u4ee3\u7684 NVIDIA GPU \u9032\u884c\u6700\u4f73\u5316\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=MGIwZTQ0ZDdhMTgxYTBlMmEzZDE5OTczN2ZlZmEzNGFfc3I3T2U0UzNCOGdjd1ZHcktHd1hkd1RpcXlLbkxrU1FfVG9rZW46Vk05WWJ0a1Bob3NkYzl4bXpFc2N6anI3bktjXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u5728\u8a13\u7df4 V3 \u6a21\u578b\u6642\uff0cDeepSeek \u91cd\u65b0\u914d\u7f6e\u4e86 NVIDIA H800 GPU\uff1a<\/p>\n\n\n\n<p>\u5728 132 \u500b\u4e32\u6d41\u8655\u7406\u5668\u6838\u5fc3\u4e2d\uff0c\u6709 20 \u500b\u5206\u914d\u7d66\u4f3a\u670d\u5668\u9593\u7684\u901a\u8a0a\uff0c\u4e3b\u8981\u7528\u65bc\u8cc7\u6599\u58d3\u7e2e\u8207\u89e3\u58d3\u7e2e\uff0c\u4ee5\u7a81\u7834\u8655\u7406\u5668\u7684\u9023\u7dda\u9650\u5236\uff0c\u63d0\u9ad8\u4ea4\u6613\u8655\u7406\u901f\u5ea6\u3002<\/p>\n\n\n\n<p>\u70ba\u4e86\u6700\u5927\u5316\u6548\u80fd\uff0cDeepSeek \u9084\u900f\u904e\u984d\u5916\u7684\u7d30\u7c92\u5ea6\u7dda\u7a0b\/\u7dda\u7a0b\u675f\u5c64\u7d1a\u8abf\u6574\uff0c\u5be6\u73fe\u4e86\u5148\u9032\u7684\u6d41\u6c34\u904b\u7b97\u6f14\u7b97\u6cd5\u3002<\/p>\n\n\n\n<p>\u9019\u4e9b\u512a\u5316\u9060\u9060\u8d85\u8d8a\u4e86\u50b3\u7d71 CUDA \u958b\u767c\u7684\u6c34\u6e96\uff0c\u4f46\u7dad\u8b77\u8d77\u4f86\u537b\u6975\u70ba\u56f0\u96e3\u3002\u7136\u800c\uff0c\u6b63\u662f\u9019\u7a2e\u512a\u5316\u6c34\u5e73\u5145\u5206\u5c55\u73fe\u4e86 DeepSeek \u5718\u968a\u5353\u8d8a\u7684\u6280\u8853\u80fd\u529b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=MDk2ZDMyM2IzOGU5OWNmN2JhOTY2ZDZmMjhkOWYwZGFfUnFiV0hvbnQ0ZUFHSHg3WHpyMW5jYTRvMURPM1pDSTZfVG9rZW46QnZVNWJyUzBDb2FWeE54Ym4ybGNZNXlnbmFnXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>V3 \u8ad6\u6587\u7279\u5225\u63d0\u5230 PTX \u7684\u8a73\u7d30\u8cc7\u8a0a<\/p>\n\n\n\n<p>\u9019\u662f\u56e0\u70ba\u5728\u5168\u7403 GPU \u77ed\u7f3a\u548c\u7f8e\u570b\u9650\u5236\u7684\u96d9\u91cd\u58d3\u529b\u4e0b\uff0cDeepSeek \u7b49\u516c\u53f8\u4e0d\u5f97\u4e0d\u5c0b\u6c42\u5275\u65b0\u7684\u89e3\u6c7a\u65b9\u6848\u3002<\/p>\n\n\n\n<p>\u5e78\u904b\u7684\u662f\uff0c\u4ed6\u5011\u5728\u9019\u65b9\u9762\u5df2\u53d6\u5f97\u91cd\u5927\u7a81\u7834\u3002<\/p>\n\n\n\n<p>\u4e00\u4f4d\u958b\u767c\u4eba\u54e1\u8a8d\u70ba\u300c\u4f4e\u968e GPU \u7a0b\u5f0f\u8a2d\u8a08\u662f\u6b63\u78ba\u7684\u65b9\u5411\u3002\u512a\u5316\u7a0b\u5ea6\u8d8a\u9ad8\uff0c\u6210\u672c\u5c31\u8d8a\u4f4e\uff0c\u4e5f\u5c31\u662f\u53ef\u4ee5\u7528\u65bc\u5176\u4ed6\u9032\u5ea6\u7684\u6548\u80fd\u9810\u7b97\uff0c\u4e0d\u9700\u8981\u984d\u5916\u652f\u51fa\"\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=MmEzYzA5ZTVmNjE4ZTlhMWE0NWU1ZTgyZTA2NmUxMDJfUWdNb21QeEFtUWlFSFA1aGFWZEZJMzlUNjdPT3J5NXRfVG9rZW46RWtaaGJ2UlBHbzk2VWF4TmxkeGNPeGdKblJnXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u9019\u4e00\u7a81\u7834\u5c0d\u5e02\u5834\u7522\u751f\u4e86\u91cd\u5927\u5f71\u97ff\uff0c\u90e8\u5206\u6295\u8cc7\u8005\u8a8d\u70ba\u65b0\u6a21\u5f0f\u5c07\u964d\u4f4e\u5e02\u5834\u5c0d\u9ad8\u6548\u80fd\u786c\u9ad4\u7684\u9700\u6c42\uff0c\u53ef\u80fd\u5f71\u97ff NVIDIA \u7b49\u516c\u53f8\u7684\u92b7\u552e\u696d\u7e3e\u3002<\/p>\n\n\n\n<p>\u4e0d\u904e\uff0c\u5305\u62ec Intel \u524d\u57f7\u884c\u9577 Pat Gelsinger \u5728\u5167\u7684\u696d\u754c\u8cc7\u6df1\u4eba\u58eb\u8a8d\u70ba\uff0cAI \u61c9\u7528\u7a0b\u5f0f\u53ef\u4ee5\u5145\u5206\u5229\u7528\u6240\u6709\u53ef\u7528\u7684\u904b\u7b97\u80fd\u529b\u3002<\/p>\n\n\n\n<p>Gelsinger \u5c07 DeepSeek \u7684\u9019\u9805\u7a81\u7834\u8996\u70ba\u5728\u5927\u773e\u5e02\u5834\u7684\u4f4e\u6210\u672c\u88dd\u7f6e\u4e2d\u5d4c\u5165 AI \u529f\u80fd\u7684\u65b0\u65b9\u6cd5\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=NzgzZjM2ZTVlOWM0OWI1MDE5OTI1NTQwNWRjYTI5Y2NfZ25sc2tPNFJ1UHZwemp1WEVlclU1cloxZXI5aHJMbEZfVG9rZW46SHlGTGJnNHpHbzNzbnd4bkxPQ2N4T0RyblZkXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"PTX_and_CUDA\"><\/span>PTX \u548c CUDA<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>\u90a3\u9ebc\uff0cDeepSeek \u7684\u51fa\u73fe\u662f\u5426\u610f\u5473\u8457\u958b\u767c\u5c16\u7aef LLM \u4e0d\u518d\u9700\u8981\u5927\u578b GPU \u96c6\u7fa4\uff1f<\/p>\n\n\n\n<p>Google \u5728\u8a08\u7b97\u8cc7\u6e90\u4e0a\u7684\u9f90\u5927\u6295\u8cc7\u662f\u5426\u6703\u3001 <a href=\"https:\/\/openai.com\/\" target=\"_blank\" rel=\"noreferrer noopener nofollow\">OpenAI<\/a>, Meta \u548c xAI \u6700\u7d42\u6703\u767d\u8cbb\u55ce\uff1fAI \u958b\u767c\u4eba\u54e1\u7684\u666e\u904d\u5171\u8b58\u662f\uff0c\u60c5\u6cc1\u4e26\u975e\u5982\u6b64\u3002<\/p>\n\n\n\n<p>\u4f46\u53ef\u4ee5\u80af\u5b9a\u7684\u662f\uff0c\u5728\u8cc7\u6599\u8655\u7406\u548c\u6f14\u7b97\u6cd5\u6700\u4f73\u5316\u65b9\u9762\u4ecd\u6709\u5de8\u5927\u7684\u6f5b\u529b\u53ef\u6316\uff0c\u672a\u4f86\u4e00\u5b9a\u6703\u51fa\u73fe\u66f4\u591a\u5275\u65b0\u7684\u6700\u4f73\u5316\u65b9\u6cd5\u3002<\/p>\n\n\n\n<p>\u7531\u65bc DeepSeek \u7684 V3 \u6a5f\u578b\u5df2\u958b\u653e\u539f\u59cb\u78bc\uff0c\u5176\u6280\u8853\u5831\u544a\u4e2d\u5df2\u8a73\u7d30\u63ed\u9732\u4e86\u76f8\u95dc\u7d30\u7bc0\u3002<\/p>\n\n\n\n<p>\u8a72\u5831\u544a\u8a18\u9304\u4e86 DeepSeek \u9032\u884c\u7684\u6df1\u5c64\u6b21\u5e95\u5c64\u512a\u5316\u3002\u7c21\u800c\u8a00\u4e4b\uff0c\u512a\u5316\u7a0b\u5ea6\u53ef\u4ee5\u6982\u62ec\u70ba\u300c\u4ed6\u5011\u5f9e\u5e95\u5c64\u91cd\u5efa\u4e86\u6574\u500b\u7cfb\u7d71\u300d\u3002<\/p>\n\n\n\n<p>\u5982\u4e0a\u6240\u8ff0\uff0c\u5728\u4f7f\u7528 H800 GPU \u8a13\u7df4 V3 \u6642\uff0cDeepSeek \u81ea\u8a02\u4e86 GPU \u7684\u6838\u5fc3\u904b\u7b97\u55ae\u5143\uff08\u6d41\u5f0f\u591a\u8655\u7406\u5668\u6838\u5fc3\uff0c\u6216 SM\uff09\uff0c\u4ee5\u6eff\u8db3\u7279\u5b9a\u9700\u6c42\u3002<\/p>\n\n\n\n<p>\u5728\u7e3d\u5171 132 \u500b SM \u4e2d\uff0c\u4ed6\u5011\u7279\u5225\u5206\u914d\u4e86 20 \u500b\u4f86\u8655\u7406\u4f3a\u670d\u5668\u9593\u7684\u901a\u8a0a\u4efb\u52d9\uff0c\u800c\u975e\u904b\u7b97\u4efb\u52d9\u3002<\/p>\n\n\n\n<p>\u9019\u7a2e\u5ba2\u88fd\u5316\u662f\u5728 PTX (\u5e73\u884c\u7dda\u7a0b\u57f7\u884c) \u5c64\u7d1a\u5b8c\u6210\uff0c\u4e5f\u5c31\u662f NVIDIA GPU \u7684\u4f4e\u968e\u6307\u4ee4\u96c6\u3002<\/p>\n\n\n\n<p>PTX \u4ee5\u63a5\u8fd1\u65bc\u532f\u7de8\u8a9e\u8a00\u7684\u5c64\u7d1a\u57f7\u884c\uff0c\u4e26\u53ef\u9032\u884c\u5fae\u7d30\u7684\u6700\u4f73\u5316\uff0c\u4f8b\u5982\u66ab\u5b58\u5668\u5206\u914d\u548c\u7dda\u7a0b\/\u7dda\u7a0b\u675f\u5c64\u7d1a\u7684\u8abf\u6574\u3002\u7136\u800c\uff0c\u9019\u7a2e\u7cbe\u7d30\u63a7\u5236\u65e2\u8907\u96dc\u53c8\u96e3\u65bc\u7dad\u8b77\u3002<\/p>\n\n\n\n<p>\u9019\u5c31\u662f\u958b\u767c\u4eba\u54e1\u901a\u5e38\u559c\u6b61\u4f7f\u7528\u9ad8\u968e\u7a0b\u5f0f\u8a9e\u8a00\uff08\u4f8b\u5982 CUDA\uff09\u7684\u539f\u56e0\uff0c\u56e0\u70ba CUDA \u53ef\u4ee5\u70ba\u5927\u591a\u6578\u5e73\u884c\u7a0b\u5f0f\u8a2d\u8a08\u4efb\u52d9\u63d0\u4f9b\u8db3\u5920\u7684\u6548\u80fd\u6700\u4f73\u5316\uff0c\u4e26\u4e14\u7701\u53bb\u4f4e\u968e\u6700\u4f73\u5316\u7684\u9700\u6c42\u3002<\/p>\n\n\n\n<p>\u7136\u800c\uff0c\u7576\u9700\u8981\u5c07 GPU \u8cc7\u6e90\u7684\u6548\u7387\u767c\u63ee\u5230\u6975\u81f4\uff0c\u4e26\u9054\u5230\u7279\u5b9a\u7684\u6700\u4f73\u5316\u8981\u6c42\u6642\uff0c\u958b\u767c\u4eba\u54e1\u5c31\u5fc5\u9808\u6c42\u52a9 PTX\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><span class=\"ez-toc-section\" id=\"However_the_technical_barriers_remain\"><\/span>\u7136\u800c\uff0c\u6280\u8853\u969c\u7919\u4ecd\u7136\u5b58\u5728<span class=\"ez-toc-section-end\"><\/span><\/h2>\n\n\n\n<p>\u5c0d\u6b64\uff0c\u7db2\u53cb Ian Cutress \u8868\u793a\uff1a\u300cDeepseek \u4f7f\u7528 PTX \u4e26\u4e0d\u80fd\u6d88\u9664 CUDA \u7684\u6280\u8853\u969c\u7919\u300d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=YTFkNWFkMTNiYjQzNDZiMDI3ZmYxYjA3MzExYjE1MGRfemNRaFdmM1R4MTMwUWVWTUxxbHN2SjZYNEhvazBrZlNfVG9rZW46SFlEU2IwNEd3b29kMGl4cmVaOGNTcHFZbmxjXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>CUDA \u662f\u4e00\u7a2e\u9ad8\u968e\u8a9e\u8a00\u3002\u5b83\u8b93\u4f7f\u7528 NVIDIA GPU \u958b\u767c\u51fd\u5f0f\u5eab\u548c\u4ecb\u9762\u8b8a\u5f97\u66f4\u5bb9\u6613\uff0c\u4e26\u652f\u63f4\u5feb\u901f\u8fed\u4ee3\u958b\u767c\u3002<\/p>\n\n\n\n<p>CUDA \u53ef\u900f\u904e\u5fae\u8abf\u5e95\u5c64\u7a0b\u5f0f\u78bc (\u5373 PTX) \u4f86\u512a\u5316\u6548\u80fd\uff0c\u800c\u4e14\u57fa\u672c\u7a0b\u5f0f\u5eab\u4e5f\u5df2\u7d93\u5b8c\u6210\u3002\u76ee\u524d\u5927\u591a\u6578\u751f\u7522\u7d1a\u8edf\u9ad4\u90fd\u662f\u5efa\u7f6e\u5728 CUDA \u4e0a\u3002<\/p>\n\n\n\n<p>PTX \u66f4\u985e\u4f3c\u65bc GPU \u53ef\u76f4\u63a5\u7406\u89e3\u7684\u5f59\u7de8\u8a9e\u8a00\u3002\u5b83\u5728\u4f4e\u968e\u904b\u4f5c\uff0c\u4e26\u5141\u8a31\u5fae\u968e\u6700\u4f73\u5316\u3002<\/p>\n\n\n\n<p>\u5982\u679c\u60a8\u9078\u64c7\u5728 PTX \u4e2d\u7de8\u7a0b\uff0c\u9019\u8868\u793a\u4e0a\u8ff0\u5167\u5efa\u7684 CUDA \u51fd\u5f0f\u5eab\u90fd\u7121\u6cd5\u4f7f\u7528\u3002\u9019\u662f\u4e00\u9805\u975e\u5e38\u7e41\u7463\u7684\u5de5\u4f5c\uff0c\u9700\u8981\u6df1\u5165\u77ad\u89e3\u786c\u9ad4\u548c\u57f7\u884c\u6642\u554f\u984c\u7684\u5c08\u696d\u77e5\u8b58\u3002<\/p>\n\n\n\n<p>\u4f46\u662f\uff0c\u5982\u679c\u958b\u767c\u4eba\u54e1\u5b8c\u5168\u77ad\u89e3\u4ed6\u5011\u6b63\u5728\u505a\u4ec0\u9ebc\uff0c\u4ed6\u5011\u78ba\u5be6\u53ef\u4ee5\u5728\u57f7\u884c\u6642\u9054\u5230\u66f4\u597d\u7684\u6548\u80fd\u548c\u6700\u4f73\u5316\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/rmy9ct2fln.feishu.cn\/space\/api\/box\/stream\/download\/asynccode\/?code=MGU3N2MyY2Y5NDg0MzAxYjkzNzI4MDg3YTRjN2JiNjJfSTI4Um5wZkxwamJMNjRLdmx6TnFmcTlhVDhLbTEyYlhfVG9rZW46SVpVcWJ4TmRtbzdYRjF4RFk5SWN4OWdjbmRlXzE3MzgxNTE4NjQ6MTczODE1NTQ2NF9WNA\" alt=\"\"\/><\/figure>\n\n\n\n<p>\u76ee\u524d\uff0cNVIDIA \u751f\u614b\u7cfb\u7d71\u7684\u4e3b\u6d41\u4ecd\u662f\u4f7f\u7528 CUDA\u3002<\/p>\n\n\n\n<p>\u60f3\u8981\u5f9e\u8a08\u7b97\u8ca0\u8f09\u4e2d\u7372\u5f97\u984d\u5916 10-20% \u6548\u80fd\u6216\u96fb\u6e90\u6548\u7387\u7684\u958b\u767c\u4eba\u54e1\uff0c\u4f8b\u5982\u5728\u96f2\u7aef\u90e8\u7f72\u6a21\u578b\u4e26\u51fa\u552e\u4ee3\u5e63\u670d\u52d9\u7684\u516c\u53f8\uff0c\u78ba\u5be6\u5df2\u5f9e CUDA \u5c64\u7d1a\u6700\u4f73\u5316\u81f3 PTX \u5c64\u7d1a\u3002\u4ed6\u5011\u9858\u610f\u6295\u5165\u6642\u9593\uff0c\u56e0\u70ba\u9577\u9060\u800c\u8a00\u9019\u662f\u503c\u5f97\u7684\u3002<\/p>\n\n\n\n<p>\u61c9\u8a72\u6ce8\u610f\u7684\u662f\uff0cPTX \u901a\u5e38\u662f\u91dd\u5c0d\u7279\u5b9a\u786c\u9ad4\u6a5f\u578b\u6700\u4f73\u5316\u7684\uff0c\u9664\u975e\u7279\u5225\u7de8\u5beb\u9069\u61c9\u908f\u8f2f\uff0c\u5426\u5247\u5f88\u96e3\u5728\u4e0d\u540c\u786c\u9ad4\u4e4b\u9593\u79fb\u690d\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u624b\u52d5\u8abf\u6574\u8a08\u7b97\u6838\u5fc3\u9700\u8981\u6975\u5927\u7684\u6bc5\u529b\u3001\u52c7\u6c23\uff0c\u4ee5\u53ca\u4fdd\u6301\u51b7\u975c\u7684\u7279\u6b8a\u80fd\u529b\uff0c\u56e0\u70ba\u7a0b\u5f0f\u53ef\u80fd\u6bcf 5,000 \u500b\u9031\u671f\u5c31\u6703\u767c\u751f\u4e00\u6b21\u8a18\u61b6\u9ad4\u5b58\u53d6\u932f\u8aa4\u3002<\/p>\n\n\n\n<p>\u7576\u7136\uff0c\u5c0d\u65bc\u90a3\u4e9b\u78ba\u5be6\u9700\u8981 PTX \u7684\u60c5\u6cc1\uff0c\u4ee5\u53ca\u90a3\u4e9b\u6709\u8db3\u5920\u85aa\u8cc7\u8655\u7406\u9019\u4e9b\u554f\u984c\u7684\u958b\u767c\u4eba\u54e1\uff0c\u6211\u5011\u8868\u793a\u5145\u5206\u7684\u7406\u89e3\u8207\u5c0a\u91cd\u3002<\/p>\n\n\n\n<p>\u5c0d\u65bc\u6240\u6709\u5176\u4ed6\u958b\u767c\u4eba\u54e1\uff0c\u5efa\u8b70\u7e7c\u7e8c\u4f7f\u7528 CUDA \u6216\u5176\u4ed6\u57fa\u65bc CUDA (\u6216 MLIR) \u7684\u9032\u968e\u8b8a\u9ad4\u3002<\/p>","protected":false},"excerpt":{"rendered":"<p>DeepSeek V3 \u8ad6\u6587\u8a73\u7d30\u5167\u5bb9\uff1a\u5982\u4f55\u7e5e\u904e CUDA \u7684\u58df\u65b7\uff01DeepSeek\u6700\u8fd1\u767c\u5e03\u7684\u5169\u6b3e\u6a21\u578bDeepSeek-V3\u548cDeepSeek-R1\uff0c\u4ee5\u66f4\u4f4e\u7684\u6210\u672c\u5be6\u73fe\u4e86\u5ab2\u7f8eOpenAI\u540c\u985e\u6a21\u578b\u7684\u6027\u80fd\u3002\u6839\u64da\u5916\u570b\u5a92\u9ad4\u5831\u5c0e\uff0c\u5728\u77ed\u77ed\u5169\u500b\u6708\u5167\uff0c\u4ed6\u5011\u5728\u4e00\u500b\u7531 2,048 \u500b\u96c6\u7fa4\u7d44\u6210\u7684\u6a21\u578b\u4e0a\uff0c\u8a13\u7df4\u51fa\u4e86\u4e00\u500b\u64c1\u6709 6,710 \u5104\u500b\u53c3\u6578\u7684 MoE \u8a9e\u8a00\u6a21\u578b...<\/p>","protected":false},"author":2,"featured_media":684,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_kadence_starter_templates_imported_post":false,"_kad_post_transparent":"","_kad_post_title":"","_kad_post_layout":"","_kad_post_sidebar_id":"","_kad_post_content_style":"","_kad_post_vertical_padding":"","_kad_post_feature":"","_kad_post_feature_position":"","_kad_post_header":false,"_kad_post_footer":false,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-710","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"_links":{"self":[{"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/posts\/710","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/comments?post=710"}],"version-history":[{"count":1,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/posts\/710\/revisions"}],"predecessor-version":[{"id":711,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/posts\/710\/revisions\/711"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/media\/684"}],"wp:attachment":[{"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/media?parent=710"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/categories?post=710"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/janusai.pro\/zh\/wp-json\/wp\/v2\/tags?post=710"}],"curies":[{"name":"\u53ef\u6fd5\u6027\u7c89\u5291","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}