2.2 GPU\u9a45\u52d5\u8207\u76f8\u4f9d\u5eab\u5b89\u88dd<\/h3>\n

\n
1. \u9a45\u52d5\u7248\u672c\u5339\u914d\uff1a\u6839\u64daGPU\u578b\u865f\u9078\u64c7\u5c0d\u61c9\u7684NVIDIA\u9a45\u52d5\u7248\u672c\u3002\u9700\u907f\u514d\u9a45\u52d5\u7248\u672c\u904e\u9ad8\u6216\u904e\u4f4e\u5c0e\u81f4\u7684\u6846\u67b6\u904b\u884c\u932f\u8aa4\uff0c\u53ef\u53c3\u8003NVIDIA\u5b98\u65b9\u6587\u4ef6\u4e2d\u7684\u9a45\u52d5-GPU\u76f8\u5bb9\u6027\u5217\u8868\u3002<\/li>\n
2. \u6838\u5fc3\u76f8\u4f9d\u5eab\u5b89\u88dd\uff1a\n
   \n
   • \u5b89\u88ddCUDA\u5de5\u5177\u5305\uff1a\u9078\u64c7\u8207\u76ee\u6a19AI\u6846\u67b6\u76f8\u5bb9\u7684CUDA\u7248\u672c\u3002\u5efa\u8b70\u4f7f\u7528\u65e5\u672c\u672c\u5730\u93e1\u50cf\u6e90\uff08\u5982\u6771\u4eac\u5de5\u696d\u5927\u5b78\u93e1\u50cf\u6e90\uff09\uff0c\u63d0\u5347\u4e0b\u8f09\u901f\u5ea6\u3002<\/li>\n
   • \u5b89\u88ddcuDNN\uff1a\u4e0b\u8f09\u8207\u5df2\u5b89\u88ddCUDA\u5de5\u5177\u5305\u7248\u672c\u5339\u914d\u7684cuDNN\uff0c\u4e26\u6b63\u78ba\u914d\u7f6e\u74b0\u5883\u8b8a\u6578\u3002<\/li>\n<\/ul>\n<\/li>\n
   • \u5b89\u88dd\u9a57\u8b49\uff1a\u5728\u7d42\u7aef\u57f7\u884c nvidia-smi<\/code> \u6307\u4ee4\u3002\u82e5\u6307\u4ee4\u8f38\u51fa\u4e2d\u6b63\u78ba\u986f\u793aGPU\u578b\u865f\u3001\u9a45\u52d5\u7248\u672c\u53caCUDA\u7248\u672c\uff0c\u5247\u8868\u793a\u9a45\u52d5\u5b89\u88dd\u6210\u529f\u3002<\/li>\n<\/ol>\n

     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n

     \u4e09\u3001\u65e5\u672cGPU\u4f3a\u670d\u5668\u4e3b\u6d41AI\u6846\u67b6\u5206\u6b65\u914d\u7f6e\u6307\u5357<\/strong><\/h2>\n

     \u672c\u7bc0\u91dd\u5c0d\u76ee\u524d\u61c9\u7528\u6700\u5ee3\u6cdb\u7684\u5169\u5927AI\u6846\u67b6\u2014\u2014TensorFlow\u8207PyTorch\uff0c\u7d50\u5408\u65e5\u672cGPU\u4f3a\u670d\u5668\u7684\u7db2\u8def\u74b0\u5883\u8207\u786c\u9ad4\u7279\u6027\uff0c\u63d0\u4f9b\u8a73\u7d30\u53ef\u843d\u5730\u7684\u914d\u7f6e\u65b9\u6848\u3002<\/p>\n

     3.1 TensorFlow GPU\u914d\u7f6e<\/h3>\n

     \n
     1. \u5b89\u88dd\u65b9\u5f0f\uff1a\u63a8\u85a6\u4f7f\u7528Conda\u800c\u975epip\u9032\u884c\u5b89\u88dd\uff0c\u900f\u904e\u74b0\u5883\u9694\u96e2\u529f\u80fd\uff0c\u907f\u514d\u4e0d\u540c\u6846\u67b6\u8207\u76f8\u4f9d\u5eab\u4e4b\u9593\u7684\u7248\u672c\u885d\u7a81\u3002\u53ef\u900f\u904e\u65e5\u672c\u672c\u5730\u93e1\u50cf\u6e90\u5b89\u88ddMiniconda\u3002<\/li>\n
     2. \u6838\u5fc3\u6b65\u9a5f\uff1a\n
        \n
        • \u5efa\u7acb\u5c08\u5c6cConda\u74b0\u5883\uff1aconda create -n tf-gpu python=3.9<\/code><\/li>\n
        • \u555f\u7528\u74b0\u5883\uff1aconda activate tf-gpu<\/code><\/li>\n
        • \u5b89\u88ddTensorFlow GPU\u7248\u672c\uff1a\u4f7f\u7528\u8207CUDA\u7248\u672c\u9069\u914d\u7684\u5b98\u65b9\u6307\u4ee4\u9032\u884c\u5b89\u88dd\uff0c\u540c\u6642\u914d\u7f6e\u65e5\u672cPyPI\u93e1\u50cf\u6e90\uff0c\u89e3\u6c7a\u6d77\u5916\u4e0b\u8f09\u901f\u5ea6\u6162\u7684\u554f\u984c\u3002<\/li>\n
        • \u914d\u7f6e\u74b0\u5883\u8b8a\u6578\uff1a\u5c07 LD_LIBRARY_PATH<\/code> \u74b0\u5883\u8b8a\u6578\u6307\u5411CUDA\u8207cuDNN\u7684\u5eab\u6a94\u6848\u76ee\u9304\u3002<\/li>\n<\/ul>\n<\/li>\n
        • \u914d\u7f6e\u9a57\u8b49\uff1a\u57f7\u884c\u4e00\u6bb5\u7c21\u55ae\u7684TensorFlow\u7a0b\u5f0f\u78bc\uff0c\u6aa2\u67e5GPU\u662f\u5426\u88ab\u6b63\u5e38\u8b58\u5225\u3002\u7bc4\u4f8b\u7a0b\u5f0f\u78bc\u5982\u4e0b\uff1a\n

          import tensorflow as tf\r\nprint(tf.config.list_physical_devices('GPU'))<\/code><\/pre>\n
          \u82e5\u7a0b\u5f0f\u78bc\u8f38\u51faGPU\u8a2d\u5099\u8cc7\u8a0a\uff0c\u5247\u8868\u793a\u914d\u7f6e\u6210\u529f\u3002<\/li>\n<\/ol>\n
          3.2 PyTorch GPU\u914d\u7f6e<\/h3>\n
          \n
          \u7248\u672c\u9078\u64c7\uff1a\u9700\u56b4\u683c\u78ba\u4fddPyTorch\u7248\u672c\u8207\u5df2\u5b89\u88dd\u7684CUDA\u7248\u672c\u76f8\u5339\u914d\u3002\u53ef\u53c3\u8003PyTorch\u5b98\u65b9\u63d0\u4f9b\u7684\u7248\u672c\u76f8\u5bb9\u6027\u5c0d\u7167\u8868\uff0c\u907f\u514d\u51fa\u73fe\u4e0d\u76f8\u5bb9\u554f\u984c\u3002<\/li>\n
          \u5b89\u88dd\u6700\u4f73\u5316\uff1a\u4f7f\u7528\u65e5\u672c\u672c\u5730\u7684PyPI\u6216Conda\u93e1\u50cf\u6e90\u52a0\u901f\u5b89\u88dd\u6d41\u7a0b\uff0c\u964d\u4f4e\u8de8\u5340\u57df\u6578\u64da\u50b3\u8f38\u5e36\u4f86\u7684\u5ef6\u9072\u3002<\/li>\n
          \u529f\u80fd\u6e2c\u8a66\uff1a\u57f7\u884c\u4e00\u500b\u7c21\u55ae\u7684\u795e\u7d93\u7db2\u8def\u8a13\u7df4\u4efb\u52d9\uff0c\u9a57\u8b49GPU\u904b\u7b97\u7684\u4f7f\u7528\u7387\u3002\u4f8b\u5982\uff0c\u5728\u6a23\u672c\u6578\u64da\u96c6\u4e0a\u8a13\u7df4\u4e00\u500b\u57fa\u790eCNN\u6a21\u578b\uff0c\u4e26\u900f\u904e nvidia-smi<\/code> \u6307\u4ee4\u76e3\u63a7GPU\u4f7f\u7528\u7387\u3002<\/li>\n<\/ol>\n
          3.3 \u5c11\u6578\u6d3eAI\u6846\u67b6\u914d\u7f6e\u6280\u5de7\uff08\u5982MindSpore\u3001MXNet\uff09<\/h3>\n
          \n
          \u9a45\u52d5\u76f8\u5bb9\u6027\u6ce8\u610f\u4e8b\u9805\uff1a\u5c11\u6578\u6d3e\u6846\u67b6\u5c0d\u9a45\u52d5\u7248\u672c\u7684\u6700\u4f4e\u8981\u6c42\u53ef\u80fd\u8207\u4e3b\u6d41\u6846\u67b6\u5b58\u5728\u5dee\u7570\uff0c\u9700\u7279\u5225\u7559\u610f\u3002<\/li>\n
          \u6545\u969c\u6392\u9664\u65b9\u6cd5\uff1a\u82e5\u5728\u65e5\u672c\u4f3a\u670d\u5668\u4e0a\u5b89\u88dd\u5c11\u6578\u6d3e\u6846\u67b6\u6642\u9047\u5230\u554f\u984c\uff0c\u53ef\u67e5\u95b1\u5b98\u65b9\u6587\u4ef6\u8207\u793e\u7fa4\u8ad6\u58c7\u3002\u540c\u6642\uff0c\u53ef\u900f\u904e\u5728\u5730\u5316\u6280\u8853\u652f\u63f4\uff0c\u5feb\u901f\u89e3\u6c7a\u554f\u984c\u3002<\/li>\n<\/ul>\n
          [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
          \u56db\u3001\u914d\u7f6e\u9a57\u8b49\u8207\u6548\u80fd\u6e2c\u8a66<\/strong><\/h2>\n
          \u5b8c\u6210AI\u6846\u67b6\u914d\u7f6e\u5f8c\uff0c\u958b\u5c55\u9a57\u8b49\u8207\u6548\u80fd\u6e2c\u8a66\u5de5\u4f5c\u81f3\u95dc\u91cd\u8981\uff0c\u9019\u80fd\u78ba\u4fdd\u6846\u67b6\u5728\u65e5\u672cGPU\u4f3a\u670d\u5668\u4e0a\u9ad8\u6548\u7a69\u5b9a\u904b\u884c\u3002<\/p>\n
          \n
          \u529f\u80fd\u9a57\u8b49\uff1a\n
          \n
          \u57f7\u884c\u6846\u67b6\u5167\u5efa\u7684\u6e2c\u8a66\u6848\u4f8b\uff0c\u9a57\u8b49\u57fa\u790e\u529f\u80fd\u662f\u5426\u6b63\u5e38\u3002<\/li>\n
          \u900f\u904e\u6846\u67b6\u5c08\u5c6c\u6307\u4ee4\u6aa2\u67e5GPU\u8b58\u5225\u72c0\u614b\uff08\u4f8b\u5982TensorFlow\u7684 tf.config.list_physical_devices('GPU')<\/code> \u6307\u4ee4\uff0c\u4ee5\u53caPyTorch\u7684 torch.cuda.is_available()<\/code> \u6307\u4ee4\uff09\u3002<\/li>\n<\/ul>\n<\/li>\n
          \u6548\u80fd\u6e2c\u8a66\uff1a\n
          \n
          \u91dd\u5c0d\u540c\u4e00\u4efb\u52d9\uff0c\u5c0d\u6bd4CPU\u8207GPU\u7684\u904b\u7b97\u901f\u5ea6\uff0c\u91cf\u5316GPU\u52a0\u901f\u5e36\u4f86\u7684\u6548\u80fd\u63d0\u5347\u6548\u679c\u3002<\/li>\n
          \u57f7\u884c\u9577\u6642\u9593\u8a13\u7df4\u4efb\u52d9\uff0c\u6e2c\u8a66\u65e5\u672c\u4f3a\u670d\u5668\u904b\u7b97\u7684\u7a69\u5b9a\u6027\uff0c\u540c\u6642\u76e3\u63a7GPU\u6eab\u5ea6\u3001\u8a18\u61b6\u9ad4\u4f54\u7528\u7b49\u95dc\u9375\u6307\u6a19\u3002<\/li>\n<\/ul>\n<\/li>\n
          \u6545\u969c\u6392\u9664\uff1a\n
          \n
          \u9a45\u52d5\u885d\u7a81\u554f\u984c\uff1a\u91cd\u65b0\u5b89\u88dd\u5339\u914d\u7248\u672c\u7684\u9a45\u52d5\uff0c\u4e26\u66f4\u65b0\u7cfb\u7d71\u76f8\u4f9d\u5eab\u3002<\/li>\n
          \u8a18\u61b6\u9ad4\u4e0d\u8db3\u554f\u984c\uff1a\u5347\u7d1a\u4f3a\u670d\u5668\u8a18\u61b6\u9ad4\u5bb9\u91cf\uff0c\u6216\u5c0d\u6a21\u578b\u9032\u884c\u6700\u4f73\u5316\u4ee5\u964d\u4f4e\u8a18\u61b6\u9ad4\u4f54\u7528\uff08\u4f8b\u5982\u63a1\u7528\u6df7\u5408\u7cbe\u5ea6\u8a13\u7df4\u6280\u8853\uff09\u3002<\/li>\n<\/ul>\n<\/li>\n<\/ol>\n
          [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
          \u4e94\u3001\u65e5\u672cGPU\u4f3a\u670d\u5668AI\u6846\u67b6\u904b\u884c\u6700\u4f73\u5316\u6280\u5de7<\/strong><\/h2>\n
          \u70ba\u5145\u5206\u767c\u63ee\u65e5\u672cGPU\u4f3a\u670d\u5668\u4e0aAI\u6846\u67b6\u7684\u6548\u80fd\uff0c\u53ef\u63a1\u53d6\u4ee5\u4e0b\u6700\u4f73\u5316\u7b56\u7565\uff1a<\/p>\n
          \n
          \u904b\u7b97\u6700\u4f73\u5316\uff1a\u958b\u555fGPU\u5e73\u884c\u904b\u7b97\u529f\u80fd\uff0c\u6839\u64da\u8a18\u61b6\u9ad4\u5bb9\u91cf\u8abf\u6574\u6279\u6b21\u5927\u5c0f\uff0c\u63a1\u7528\u6df7\u5408\u7cbe\u5ea6\u8a13\u7df4\u6280\u8853\uff0c\u63d0\u5347\u904b\u7b97\u6548\u7387\u3002<\/li>\n
          \u7db2\u8def\u6700\u4f73\u5316\uff1a\u7d81\u5b9a\u65e5\u672c\u672c\u5730DNS\u4f3a\u670d\u5668\uff0c\u7e2e\u77ed\u7db2\u57df\u540d\u7a31\u89e3\u6790\u6642\u9593\uff1b\u555f\u7528TCP\u52a0\u901f\u6280\u8853\uff0c\u63d0\u5347\u6a21\u578b\u4e0b\u8f09\u8207\u6578\u64da\u50b3\u8f38\u901f\u5ea6\u3002<\/li>\n
          \u7dad\u904b\u6700\u4f73\u5316\uff1a\u5b9a\u671f\u66f4\u65b0GPU\u9a45\u52d5\u8207\u6846\u67b6\u7248\u672c\uff0c\u4fee\u5fa9\u5df2\u77e5\u6f0f\u6d1e\u4e26\u63d0\u5347\u6548\u80fd\uff1b\u900f\u904e\u65e5\u672c\u8cc7\u6599\u4e2d\u5fc3\u7684\u6eab\u63a7\u512a\u52e2\uff0c\u5373\u6642\u76e3\u63a7GPU\u6eab\u5ea6\uff0c\u9632\u6b62\u8a2d\u5099\u904e\u71b1\u3002<\/li>\n
          \u6210\u672c\u6700\u4f73\u5316\uff1a\u6839\u64da\u5c08\u6848\u9700\u6c42\uff0c\u9748\u6d3b\u9078\u64c7\u6309\u9700\u8a08\u8cbb\u3001\u5305\u6708\u6216\u5305\u5e74\u7684\u8a08\u8cbb\u6a21\u5f0f\uff0c\u964d\u4f4eAI\u7814\u767c\u6210\u672c\u3002\u5c0d\u65bc\u9577\u671f\u5c08\u6848\uff0c\u5305\u6708\u6216\u5305\u5e74\u8a08\u8cbb\u6a21\u5f0f\u7684\u6027\u50f9\u6bd4\u66f4\u9ad8\u3002<\/li>\n<\/ol>\n
          [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
          \u516d\u3001\u5e38\u898b\u554f\u984c\u89e3\u7b54\uff08FAQ\uff09<\/strong><\/h2>\n
          \n
          \u554f\uff1a\u65e5\u672cGPU\u4f3a\u670d\u5668\u5b89\u88dd\u9a45\u52d5\u5f8c\uff0c\u63d0\u793a\u300c\u672a\u627e\u5230\u8a2d\u5099\u300d\u8a72\u5982\u4f55\u89e3\u6c7a\uff1f
          \n\u7b54\uff1a\u6aa2\u67e5\u9a45\u52d5\u7248\u672c\u8207GPU\u578b\u865f\u662f\u5426\u5339\u914d\u3002\u5728BIOS\u4e2d\u95dc\u9589\u5b89\u5168\u958b\u555f\u529f\u80fd\u5f8c\uff0c\u91cd\u65b0\u5b89\u88dd\u9a45\u52d5\u3002\u82e5\u554f\u984c\u4ecd\u672a\u89e3\u6c7a\uff0c\u53ef\u806f\u7e6b\u65e5\u672c\u4f3a\u670d\u5668\u79df\u7528\u670d\u52d9\u5546\u7684\u6280\u8853\u652f\u63f4\uff0c\u9032\u884c\u786c\u9ad4\u6aa2\u6e2c\u3002<\/li>\n
          \u554f\uff1aTensorFlow\u7121\u6cd5\u8b58\u5225GPU\uff0c\u50c5\u4f7f\u7528CPU\u9032\u884c\u904b\u7b97\uff0c\u8a72\u5982\u4f55\u8655\u7406\uff1f
          \n\u7b54\uff1a\u9a57\u8b49TensorFlow\u7248\u672c\u8207CUDA\/cuDNN\u7248\u672c\u7684\u76f8\u5bb9\u6027\uff0c\u6aa2\u67e5\u74b0\u5883\u8b8a\u6578\u662f\u5426\u914d\u7f6e\u6b63\u78ba\u3002\u5fc5\u8981\u6642\uff0c\u53ef\u5728\u5168\u65b0\u7684Conda\u74b0\u5883\u4e2d\u91cd\u65b0\u5b89\u88ddTensorFlow\u3002<\/li>\n
          \u554f\uff1a\u540c\u4e00\u53f0\u65e5\u672cGPU\u4f3a\u670d\u5668\u4e0a\uff0c\u80fd\u5426\u540c\u6642\u57f7\u884c\u591a\u500bAI\u6846\u67b6\uff1f
          \n\u7b54\uff1a\u53ef\u4ee5\u3002\u900f\u904eConda\u70ba\u4e0d\u540c\u6846\u67b6\u5efa\u7acb\u76f8\u4e92\u9694\u96e2\u7684\u904b\u884c\u74b0\u5883\uff0c\u78ba\u4fdd\u5404\u6846\u67b6\u7684\u76f8\u4f9d\u5eab\u4e0d\u6703\u767c\u751f\u885d\u7a81\u3002<\/li>\n
          \u554f\uff1a\u65e5\u672c\u4f3a\u670d\u5668\u983b\u5bec\u4e0d\u8db3\uff0c\u5f71\u97ffAI\u6a21\u578b\u8a13\u7df4\u600e\u9ebc\u8fa6\uff1f
          \n\u7b54\uff1a\u5347\u7d1a\u4f3a\u670d\u5668\u983b\u5bec\u65b9\u6848\uff0c\u9078\u64c7\u642d\u8f09BGP\u591a\u7dda\u983b\u5bec\u7684\u65e5\u672cGPU\u4f3a\u670d\u5668\uff0c\u4fdd\u969c\u6578\u64da\u50b3\u8f38\u7a69\u5b9a\u6027\u3002\u540c\u6642\uff0c\u53ef\u5c0d\u8a13\u7df4\u6578\u64da\u9032\u884c\u58d3\u7e2e\u8655\u7406\uff0c\u6216\u63a1\u7528\u672c\u5730\u6578\u64da\u5feb\u53d6\u6280\u8853\uff0c\u6e1b\u5c11\u983b\u5bec\u4f54\u7528\u3002<\/li>\n<\/ol>\n
          [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
          \u4e03\u3001\u7e3d\u7d50<\/strong><\/h2>\n
          \u70ba\u7279\u5b9aAI\u6846\u67b6\u914d\u7f6e\u65e5\u672cGPU\u4f3a\u670d\u5668\uff0c\u4e3b\u8981\u5305\u542b\u56db\u5927\u6838\u5fc3\u6b65\u9a5f\uff1a\u4f3a\u670d\u5668\u9078\u578b\u3001\u7cfb\u7d71\u74b0\u5883\u5efa\u7f6e\u3001\u6846\u67b6\u5b89\u88dd\u914d\u7f6e\u4ee5\u53ca\u9a57\u8b49\u6700\u4f73\u5316\u3002\u65e5\u672cGPU\u4f3a\u670d\u5668\u6191\u85c9\u4f4e\u5ef6\u9072\u3001\u9ad8\u5408\u898f\u6027\u3001\u5728\u5730\u5316\u652f\u63f4\u5b8c\u5584\u7b49\u512a\u52e2\uff0c\u70ba\u4e9e\u592a\u5730\u5340\u7684AI\u61c9\u7528\u63d0\u4f9b\u4e86\u53ef\u9760\u7684\u904b\u7b97\u652f\u6490\u3002\u9075\u5faa\u672c\u6587\u7684\u6280\u8853\u6307\u5357\uff0c\u4f60\u80fd\u5920\u9ad8\u6548\u5b8c\u6210\u914d\u7f6e\u6d41\u7a0b\uff0c\u898f\u907f\u5e38\u898b\u554f\u984c\uff0c\u6700\u5927\u5316\u767c\u63eeAI\u6846\u67b6\u7684\u6548\u80fd\u3002\u7121\u8ad6\u4f60\u5f9e\u4e8b\u5b78\u8853\u7814\u7a76\u9084\u662f\u5de5\u696d\u61c9\u7528\uff0c\u70baAI\u6846\u67b6\u91cf\u8eab\u6253\u9020\u7684\u65e5\u672cGPU\u4f3a\u670d\u5668\u914d\u7f6e\u65b9\u6848<\/strong>\uff0c\u90fd\u5c07\u986f\u8457\u63d0\u5347\u4f60\u7684\u7814\u767c\u6548\u7387\u3002\u66f4\u591aAI\u8207\u4f3a\u670d\u5668\u914d\u7f6e\u76f8\u95dc\u7684\u6280\u8853\u6307\u5357\uff0c\u656c\u8acb\u95dc\u6ce8\u672c\u7db2\u7ad9\u3002<\/p>\n
          [\/vc_column_text][\/vc_column][\/vc_row]<\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"
          [vc_row el_class=”blog-detail-section”][vc_ […]<\/p>\n
          Read More…<\/a><\/p>\n","protected":false},"author":1,"featured_media":29575,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[8116],"tags":[9830,9831,9486,8159],"class_list":["post-29581","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-japan-dedicated-server-tc","tag-ai-framework-deployment","tag-apac-ai-server","tag-gpu-server-configuration","tag-japan-server-hosting-tc"],"acf":[],"yoast_head":"\n