{"id":5423,"date":"2023-04-06T11:04:19","date_gmt":"2023-04-06T03:04:19","guid":{"rendered":"https:\/\/ixyzero.com\/blog\/?p=5423"},"modified":"2023-04-06T11:04:19","modified_gmt":"2023-04-06T03:04:19","slug":"cpu%e5%92%8cgpu%e7%9a%84%e5%8c%ba%e5%88%ab","status":"publish","type":"post","link":"https:\/\/ixyzero.com\/blog\/archives\/5423.html","title":{"rendered":"CPU\u548cGPU\u7684\u533a\u522b"},"content":{"rendered":"\n

=Start=<\/p>\n\n\n\n

\u7f18\u7531\uff1a<\/h4>\n\n\n\n

\u81ea\u4ece chatGPT \u6a2a\u7a7a\u51fa\u4e16\u4e4b\u540e\uff0c\u5173\u4e8e AI \u7684\u8ba8\u8bba\u5c42\u51fa\u4e0d\u7a77\uff0c\u4e3a\u4e86\u907f\u514d\u4fe1\u606f\u8131\u8282\uff0c\u9700\u8981\u4e3b\u52a8\u53bb\u5b66\u4e60\u4e86\u89e3\u76f8\u5173\u77e5\u8bc6\u3002\u5728\u770b\u4e86\u4e00\u4e9b\u6587\u7ae0\u4e4b\u540e\uff0c\u53d1\u73b0\u6211\u6709\u592a\u591a\u7684\u57fa\u7840\u77e5\u8bc6\u8981\u8865\u5145\uff0c\u53ea\u80fd\u4e00\u70b9\u4e00\u70b9\u6765\uff0c\u5148\u4ece\u57fa\u7840\u7684\u5f00\u59cb\u3002\u8fd9\u6b21\u5148\u5b66\u4e60\u6574\u7406\u4e00\u4e0bCPU\u548cGPU\u7684\u533a\u522b\uff0c\u624d\u597d\u53bb\u4e86\u89e3\u4e3a\u4ec0\u4e48\u6df1\u5ea6\u5b66\u4e60\u7b49\u673a\u5668\u5b66\u4e60\u7b97\u6cd5\u7684\u4ecb\u7ecd\u57fa\u672c\u79bb\u4e0d\u5f00GPU\uff0c\u4e3a\u4ec0\u4e48NVIDIA\u82f1\u4f1f\u8fbe\u4f1a\u548cGPT\u4ea7\u751f\u5982\u6b64\u7d27\u5bc6\u7684\u8054\u7cfb\uff1f<\/p>\n\n\n\n

\u6b63\u6587\uff1a<\/h4>\n\n\n\n

\u53c2\u8003\u89e3\u7b54\uff1a<\/h5>\n\n\n\n
CPU\u548cGPU\u90fd\u662f\u5904\u7406\u5668\uff0c\u4e24\u8005\u6709\u4ec0\u4e48\u4e0d\u540c<\/h5>\n\n\n\n

\u8bbe\u8ba1\u76ee\u6807\u4e0d\u540c\u5bfc\u81f4\u63a5\u4e0b\u6765\u7684\u4e00\u7cfb\u5217\u4e0d\u540c\u3002<\/mark><\/strong><\/p>\n\n\n\n

CPU\u548cGPU\u4e4b\u6240\u4ee5\u5927\u4e0d\u76f8\u540c\uff0c\u662f\u7531\u4e8e\u5176\u8bbe\u8ba1\u76ee\u6807\u7684\u4e0d\u540c\uff0c\u5b83\u4eec\u5206\u522b\u9488\u5bf9\u4e86\u4e24\u79cd\u4e0d\u540c\u7684\u5e94\u7528\u573a\u666f\u3002CPU\u9700\u8981\u5f88\u5f3a\u7684\u901a\u7528\u6027<\/strong>\u6765\u5904\u7406\u5404\u79cd\u4e0d\u540c\u7684\u6570\u636e\u7c7b\u578b\uff0c\u540c\u65f6\u53c8\u8981\u903b\u8f91\u5224\u65ad\u53c8\u4f1a\u5f15\u5165\u5927\u91cf\u7684\u5206\u652f\u8df3\u8f6c\u548c\u4e2d\u65ad\u7684\u5904\u7406\u3002\u8fd9\u4e9b\u90fd\u4f7f\u5f97CPU\u7684\u5185\u90e8\u7ed3\u6784\u5f02\u5e38\u590d\u6742\u3002<\/mark>\u800cGPU\u9762\u5bf9\u7684\u5219\u662f\u7c7b\u578b\u9ad8\u5ea6\u7edf\u4e00\u7684\u3001\u76f8\u4e92\u65e0\u4f9d\u8d56\u7684\u5927\u89c4\u6a21\u6570\u636e\u548c\u4e0d\u9700\u8981\u88ab\u6253\u65ad\u7684\u7eaf\u51c0\u7684\u8ba1\u7b97\u73af\u5883<\/strong>\u3002<\/p>\n\n\n\n

GPU\u91c7\u7528\u4e86\u6570\u91cf\u4f17\u591a\u7684\u8ba1\u7b97\u5355\u5143\u548c\u8d85\u957f\u7684\u6d41\u6c34\u7ebf\uff0c\u4f46\u53ea\u6709\u975e\u5e38\u7b80\u5355\u7684\u63a7\u5236\u903b\u8f91<\/mark>\u5e76\u7701\u53bb\u4e86Cache\u3002\u800cCPU\u4e0d\u4ec5\u88abCache\u5360\u636e\u4e86\u5927\u91cf\u7a7a\u95f4\uff0c\u800c\u4e14\u8fd8\u6709\u590d\u6742\u7684\u63a7\u5236\u903b\u8f91\u548c\u8bf8\u591a\u4f18\u5316\u7535\u8def<\/mark>\uff0c\u76f8\u6bd4\u4e4b\u4e0b\u8ba1\u7b97\u80fd\u529b\u53ea\u662fCPU\u5f88\u5c0f\u7684\u4e00\u90e8\u5206\u3002<\/p>\n\n\n\n

\u7406\u89e3 GPU \u548c CPU \u4e4b\u95f4\u533a\u522b\u7684\u4e00\u79cd\u7b80\u5355\u65b9\u5f0f\u662f\u6bd4\u8f83\u5b83\u4eec\u5982\u4f55\u5904\u7406\u4efb\u52a1\u3002CPU \u7531\u4e13\u4e3a\u987a\u5e8f\u4e32\u884c\u5904\u7406<\/strong>\u800c\u4f18\u5316\u7684\u51e0\u4e2a\u6838\u5fc3\u7ec4\u6210\uff0c\u800c GPU \u5219\u62e5\u6709\u4e00\u4e2a\u7531\u6570\u4ee5\u5343\u8ba1\u7684\u66f4\u5c0f\u3001\u66f4\u9ad8\u6548\u7684\u6838\u5fc3\uff08\u4e13\u4e3a\u540c\u65f6\u5904\u7406\u591a\u91cd\u4efb\u52a1\u800c\u8bbe\u8ba1\uff09\u7ec4\u6210\u7684\u5927\u89c4\u6a21\u5e76\u884c\u8ba1\u7b97<\/strong>\u67b6\u6784\u3002<\/p>\n\n\n\n

CPU (Central Processing Unit\uff0c\u4e2d\u592e\u5904\u7406\u5668)\u5c31\u662f\u673a\u5668\u7684\u201c\u5927\u8111\u201d\uff0c\u662f\u5b8c\u6210\u5e03\u5c40\u8c0b\u7565\u3001\u53d1\u53f7\u65bd\u4ee4\u3001\u63a7\u5236\u884c\u52a8\u7684\u201c\u603b\u53f8\u4ee4\u5b98\u201d\u3002CPU\u7684\u7ed3\u6784\u4e3b\u8981\u5305\u62ec\u8fd0\u7b97\u5668(ALU\uff0cArithmetic and Logic Unit)\u3001\u63a7\u5236\u5355\u5143(CU\uff0cControl Unit)\u3001\u5bc4\u5b58\u5668(Register)\u3001\u9ad8\u901f\u7f13\u5b58\u5668(Cache)\u548c\u4e4b\u95f4\u901a\u8baf\u7684\u6570\u636e\u3001\u63a7\u5236\u53ca\u72b6\u6001\u7684\u603b\u7ebf\u3002<\/p>\n\n\n\n

GPU(Graphics Processing Unit\uff0c\u4e2d\u6587\u4e3a\u56fe\u5f62\u5904\u7406\u5668)\uff0c\u53c8\u79f0\u663e\u793a\u6838\u5fc3\u3001\u89c6\u89c9\u5904\u7406\u5668\u3001\u663e\u793a\u82af\u7247\u3002<\/mark>\u5c31\u5982\u5176\u540d\u5b57\u4e00\u6837\uff0cGPU\u6700\u521d\u662f\u7528\u5728\u4e2a\u4eba\u7535\u8111\u3001\u5de5\u4f5c\u7ad9\u3001\u6e38\u620f\u673a\u548c\u4e00\u4e9b\u79fb\u52a8\u8bbe\u5907(\u5982\u5e73\u677f\u7535\u8111\u3001\u667a\u80fd\u624b\u673a\u7b49)\u4e0a\u8fd0\u884c\u7ed8\u56fe\u8fd0\u7b97\u5de5\u4f5c\u7684\u5fae\u5904\u7406\u5668\u3002<\/p>\n\n\n\n

CPU \u57fa\u4e8e\u4f4e\u5ef6\u65f6\u7684\u8bbe\u8ba1(latency oriented design)\u3002<\/p>\n\n\n\n

GPU\u662f\u57fa\u4e8e\u5927\u7684\u541e\u5410\u91cf\u8bbe\u8ba1(throughput oriented design)\u3002<\/p>\n\n\n\n

\u4e0eCPU\u64c5\u957f\u903b\u8f91\u63a7\u5236\uff0c\u4e32\u884c\u7684\u8fd0\u7b97\u548c\u901a\u7528\u7c7b\u578b\u6570\u636e\u8fd0\u7b97\u4e0d\u540c\uff0cGPU\u64c5\u957f\u7684\u662f\u5927\u89c4\u6a21\u5e76\u53d1\u8ba1\u7b97\uff0c\u8fd9\u4e5f\u6b63\u662f\u5bc6\u7801\u7834\u89e3\u7b49\u6240\u9700\u8981\u7684\u3002\u6240\u4ee5GPU\u9664\u4e86\u56fe\u50cf\u5904\u7406\uff0c\u4e5f\u8d8a\u6765\u8d8a\u591a\u5730\u53c2\u4e0e\u5230\u8ba1\u7b97\u5f53\u4e2d\u6765\u3002<\/p>\n\n\n\n

\u4e00\u4e2a\u7b80\u5355\u7684\u7c7b\u6bd4\u2014\u2014\u6559\u6388vs\u5c0f\u5b66\u751f<\/h5>\n\n\n\n

CPU\u548cGPU\u7684\u6700\u5927\u4e0d\u540c\u5728\u4e8e\u67b6\u6784\uff0cCPU\u9002\u7528\u4e8e\u5e7f\u6cdb\u7684\u5e94\u7528\u573a\u666f(\u5b66\u8bc6\u6e0a\u535a)\uff0c\u53ef\u4ee5\u6267\u884c\u4efb\u610f\u7a0b\u5e8f\uff0c\u800cGPU\u5219\u4e13\u4e3a\u591a\u4efb\u52a1\u800c\u751f\uff0c\u5e76\u53d1\u80fd\u529b\u5f3a\uff0c\u5177\u4f53\u6765\u8bb2\u5c31\u662f\u591a\u6838\uff0c\u4e00\u822c\u7684CPU\u67092\u6838\u30014\u6838\u30018\u6838\u7b49\uff0c\u800cGPU\u5219\u53ef\u80fd\u4f1a\u6709\u6210\u767e\u4e0a\u5343\u6838\u3002<\/p>\n\n\n\n

GPU\u7684\u5de5\u4f5c\u5927\u90e8\u5206\u5c31\u662f\u8fd9\u6837\uff0c\u8ba1\u7b97\u91cf\u5927\uff0c\u4f46\u6ca1\u4ec0\u4e48\u6280\u672f\u542b\u91cf\uff0c\u800c\u4e14\u8981\u91cd\u590d\u5f88\u591a\u5f88\u591a\u6b21<\/strong>\u3002\u5c31\u50cf\u4f60\u6709\u4e2a\u5de5\u4f5c\u9700\u8981\u7b97\u51e0\u4ebf\u6b21\u4e00\u767e\u4ee5\u5185\u52a0\u51cf\u4e58\u9664\u4e00\u6837\uff0c\u6700\u597d\u7684\u529e\u6cd5\u5c31\u662f\u96c7\u4e0a\u51e0\u5341\u4e2a\u5c0f\u5b66\u751f\u4e00\u8d77\u7b97\uff0c\u4e00\u4eba\u7b97\u4e00\u90e8\u5206\uff0c\u53cd\u6b63\u8fd9\u4e9b\u8ba1\u7b97\u4e5f\u6ca1\u4ec0\u4e48\u6280\u672f\u542b\u91cf\uff0c\u7eaf\u7cb9\u4f53\u529b\u6d3b\u800c\u5df2\u3002\u800cCPU\u5c31\u50cf\u8001\u6559\u6388\uff0c\u79ef\u5206\u5fae\u5206\u90fd\u4f1a\u7b97\uff0c\u5c31\u662f\u5de5\u8d44\u9ad8\uff0c\u4e00\u4e2a\u8001\u6559\u6388\u8d44\u9876\u4e8c\u5341\u4e2a\u5c0f\u5b66\u751f\uff0c\u4f60\u8981\u662f\u5bcc\u58eb\u5eb7\u4f60\u96c7\u54ea\u4e2a\uff1fGPU\u5c31\u662f\u8fd9\u6837\uff0c\u7528\u5f88\u591a\u7b80\u5355\u7684\u8ba1\u7b97\u5355\u5143\u53bb\u5b8c\u6210\u5927\u91cf\u7684\u8ba1\u7b97\u4efb\u52a1\uff0c\u7eaf\u7cb9\u7684\u4eba\u6d77\u6218\u672f\u3002\u8fd9\u79cd\u7b56\u7565\u57fa\u4e8e\u4e00\u4e2a\u524d\u63d0\uff0c\u5c31\u662f\u5c0f\u5b66\u751fA\u548c\u5c0f\u5b66\u751fB\u7684\u5de5\u4f5c\u6ca1\u6709\u4ec0\u4e48\u4f9d\u8d56\u6027\uff0c\u662f\u4e92\u76f8\u72ec\u7acb\u7684<\/strong>\u3002\u5f88\u591a\u6d89\u53ca\u5230\u5927\u91cf\u8ba1\u7b97\u7684\u95ee\u9898\u57fa\u672c\u90fd\u6709\u8fd9\u79cd\u7279\u6027\uff0c\u6bd4\u5982\u4f60\u8bf4\u7684\u7834\u89e3\u5bc6\u7801\uff0c\u6316\u77ff\u548c\u5f88\u591a\u56fe\u5f62\u5b66\u7684\u8ba1\u7b97<\/strong>\u3002\u8fd9\u4e9b\u8ba1\u7b97\u53ef\u4ee5\u5206\u89e3\u4e3a\u591a\u4e2a\u76f8\u540c\u7684\u7b80\u5355\u5c0f\u4efb\u52a1\uff0c\u6bcf\u4e2a\u4efb\u52a1\u5c31\u53ef\u4ee5\u5206\u7ed9\u4e00\u4e2a\u5c0f\u5b66\u751f\u53bb\u505a\u3002\u4f46\u8fd8\u6709\u4e00\u4e9b\u4efb\u52a1\u6d89\u53ca\u5230\u201c\u6d41\u201d\u7684\u95ee\u9898\u3002\u6bd4\u5982\u4f60\u53bb\u76f8\u4eb2\uff0c\u53cc\u65b9\u770b\u7740\u987a\u773c\u624d\u80fd\u7ee7\u7eed\u53d1\u5c55\u3002\u603b\u4e0d\u80fd\u4f60\u8fd9\u8fb9\u8fd8\u6ca1\u89c1\u9762\u5462\uff0c\u90a3\u8fb9\u627e\u4eba\u628a\u8bc1\u90fd\u7ed9\u9886\u4e86\u3002\u8fd9\u79cd\u6bd4\u8f83\u590d\u6742\u7684\u95ee\u9898\u90fd\u662fCPU\u6765\u505a\u7684\u3002<\/p>\n\n\n\n

\u603b\u800c\u8a00\u4e4b\uff0cCPU\u548cGPU\u56e0\u4e3a\u6700\u521d\u7528\u6765\u5904\u7406\u7684\u4efb\u52a1\u5c31\u4e0d\u540c\uff0c\u6240\u4ee5\u8bbe\u8ba1\u4e0a\u6709\u4e0d\u5c0f\u7684\u533a\u522b\u3002\u800c\u67d0\u4e9b\u4efb\u52a1\u548cGPU\u6700\u521d\u7528\u6765\u89e3\u51b3\u7684\u95ee\u9898\u6bd4\u8f83\u76f8\u4f3c\uff0c\u6240\u4ee5\u7528GPU\u6765\u7b97\u4e86\u3002GPU\u7684\u8fd0\u7b97\u901f\u5ea6\u53d6\u51b3\u4e8e\u96c7\u4e86\u591a\u5c11\u5c0f\u5b66\u751f\uff0cCPU\u7684\u8fd0\u7b97\u901f\u5ea6\u53d6\u51b3\u4e8e\u8bf7\u4e86\u591a\u4e48\u5389\u5bb3\u7684\u6559\u6388\u3002\u6559\u6388\u5904\u7406\u590d\u6742\u4efb\u52a1\u7684\u80fd\u529b\u662f\u78be\u538b\u5c0f\u5b66\u751f\u7684\uff0c\u4f46\u662f\u5bf9\u4e8e\u6ca1\u90a3\u4e48\u590d\u6742\u7684\u4efb\u52a1\uff0c\u8fd8\u662f\u9876\u4e0d\u4f4f\u4eba\u591a<\/strong>\u3002<\/mark>\u5f53\u7136\u73b0\u5728\u7684GPU\u4e5f\u80fd\u505a\u4e00\u4e9b\u7a0d\u5fae\u590d\u6742\u7684\u5de5\u4f5c\u4e86\uff0c\u76f8\u5f53\u4e8e\u5347\u7ea7\u6210\u521d\u4e2d\u751f\u9ad8\u4e2d\u751f\u7684\u6c34\u5e73\u3002\u4f46\u8fd8\u9700\u8981CPU\u6765\u628a\u6570\u636e\u5582\u5230\u5634\u8fb9\u624d\u80fd\u5f00\u59cb\u5e72\u6d3b\uff0c\u7a76\u7adf\u8fd8\u662f\u9760CPU\u6765\u7ba1\u7684\u3002<\/p>\n\n\n\n

\u4e3a\u4ec0\u4e48GPU\u9700\u8981\u8fd9\u4e48\u591a\u6838\u5fc3\uff1f<\/h5>\n\n\n\n

\u60f3\u4e00\u60f3\u8ba1\u7b97\u673a\u4e0a\u7684\u4e00\u5f20\u56fe\u662f\u600e\u4e48\u8868\u793a\u7684\uff1f\u65e0\u975e\u5c31\u662f\u5c4f\u5e55\u4e0a\u7684\u4e00\u4e2a\u4e2a\u50cf\u7d20\u3002<\/p>\n\n\n\n

\u6211\u4eec\u9700\u8981\u4e3a\u6bcf\u4e2a\u50cf\u7d20\u8fdb\u884c\u8ba1\u7b97\uff0c\u800c\u4e14\u662f\u76f8\u540c\u7684\u8fd0\u7b97\uff0c\u5c31\u597d\u6bd4\u521a\u624d\u4f8b\u5b50\u4e2d\u7684\u5c0f\u5b66\u751f\u8ba1\u7b97\u8ba1\u52a0\u6cd5\u4e00\u6837\uff0c\u6ce8\u610f\uff0c\u5bf9\u4e8e\u5c4f\u5e55\u6765\u8bf4\u4e00\u822c\u4f1a\u6709\u4e0a\u767e\u4e07\u4e2a\u50cf\u7d20\uff0c\u5982\u679c\u6211\u4eec\u8981\u4e32\u884c\u7684\u4e3a\u6bcf\u4e00\u4e2a\u50cf\u7d20\u8fdb\u884c\u8fd0\u7b97\u6548\u7387\u5c31\u592a\u4f4e\u4e86\uff0c\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u8ba9GPU\u4e2d\u7684\u6bcf\u4e00\u4e2a\u6838\u5fc3\u53bb\u4e3a\u76f8\u5e94\u7684\u50cf\u7d20\u8fdb\u884c\u8ba1\u7b97\uff0c\u7531\u4e8eGPU\u4e2d\u6709\u5f88\u591a\u6838\u5fc3\uff0c\u56e0\u6b64\u5e76\u884c\u8ba1\u7b97\u53ef\u4ee5\u5927\u5e45\u63d0\u9ad8\u901f\u5ea6\u3002<\/p>\n\n\n\n

\u73b0\u5728\u4f60\u5e94\u8be5\u660e\u767d\u4e3a\u4ec0\u4e48GPU\u8981\u8fd9\u6837\u5de5\u4f5c\u4e86\u5427\u3002<\/p>\n\n\n\n

\u4e0d\u540c\u7684\u5de5\u4f5c\u65b9\u5f0f<\/h5>\n\n\n\n

\u5bf9CPU\u6765\u8bf4\uff0c\u4e0d\u540c\u7684\u6838\u5fc3\u53ef\u4ee5\u6267\u884c\u4e0d\u540c\u7684\u673a\u5668\u6307\u4ee4\uff0ccoreA\u5728\u8fd0\u884cword\u7ebf\u7a0b\u7684\u540c\u65f6coreB\u4e0a\u53ef\u4ee5\u8fd0\u884c\u6d4f\u89c8\u5668\u7ebf\u7a0b\uff0c\u8fd9\u5c31\u662f\u6240\u8c13\u7684\u591a\u6307\u4ee4\u591a\u6570\u636e\uff0cMIMD\uff0c(Multiple Instruction, Multiple Data)\u3002<\/p>\n\n\n\n

\u800cGPU\u5219\u4e0d\u540c\uff0cGPU\u4e0a\u7684\u8fd9\u4e9b\u6838\u5fc3\u5fc5\u987b\u6574\u9f50\u5212\u4e00\u7684\u8fd0\u884c\u76f8\u540c\u7684\u673a\u5668\u6307\u4ee4\uff0c\u53ea\u662f\u53ef\u4ee5\u64cd\u4f5c\u4e0d\u540c\u7684\u6570\u636e\uff0c\u8fd9\u5c31\u597d\u6bd4\u8fd9\u4e9b\u5c0f\u5b66\u751f\u5728\u67d0\u4e2a\u65f6\u523b\u5fc5\u987b\u90fd\u8fdb\u884c\u52a0\u6cd5\u8ba1\u7b97\uff0c\u4e0d\u540c\u7684\u5730\u65b9\u5728\u4e8e\u6709\u7684\u5c0f\u5b66\u751f\u53ef\u80fd\u9700\u8981\u8ba1\u7b971+1\uff0c\u6709\u7684\u8981\u8ba1\u7b972+6\u7b49\u7b49\uff0c\u53d8\u5316\u7684\u5730\u65b9\u4ec5\u5728\u4e8e\u64cd\u4f5c\u6570\uff0c\u8fd9\u5c31\u662f\u6240\u8c13\u7684\u5355\u6307\u4ee4\u591a\u6570\u636e\uff0cSIMD\uff0c(Single Instruction, Multiple Data)\u3002<\/p>\n\n\n\n

\u56e0\u6b64\u6211\u4eec\u53ef\u4ee5\u770b\u5230GPU\u7684\u5de5\u4f5c\u65b9\u5f0f\u548cCPU\u662f\u622a\u7136\u4e0d\u540c\u7684\u3002<\/p>\n\n\n\n

\u9664\u4e86\u8fd9\u79cd\u5de5\u4f5c\u65b9\u5f0f\u4e4b\u5916\uff0cGPU\u7684\u6307\u4ee4\u96c6\u8fd8\u975e\u5e38\u7b80\u5355\uff0c\u4e0d\u50cfCPU\u8fd9\u79cd\u590d\u6742\u7684\u5904\u7406\u5668\uff0c\u5982\u679c\u4f60\u53bb\u770bCPU\u7684\u7f16\u7a0b\u624b\u518c\u5c31\u4f1a\u53d1\u73b0\uff0cCPU\u8d1f\u8d23\u7684\u4e8b\u60c5\u975e\u5e38\u591a\uff1a\u4e2d\u65ad\u5904\u7406\u3001\u5185\u5b58\u7ba1\u7406\u3001IO\u7b49\u7b49\uff0c\u8fd9\u4e9b\u5bf9\u4e8eGPU\u6765\u8bf4\u90fd\u662f\u4e0d\u5b58\u5728\u7684\uff0c\u53ef\u4ee5\u770b\u5230GPU\u7684\u5b9a\u4f4d\u975e\u5e38\u7b80\u5355\uff0c\u5c31\u662f\u7eaf\u8ba1\u7b97\uff0cGPU\u7edd\u4e0d\u662f\u7528\u6765\u53d6\u4ee3CPU\u7684\uff0cCPU\u53ea\u662f\u628a\u4e00\u4e9bGPU\u975e\u5e38\u64c5\u957f\u7684\u4e8b\u60c5\u4ea4\u7ed9\u5b83\uff0cGPU\u4ec5\u4ec5\u662f\u7528\u6765\u5206\u62c5CPU\u5de5\u4f5c\u7684\u914d\u89d2\u3002<\/p>\n\n\n\n

GPU\u64c5\u957f\u4ec0\u4e48<\/h5>\n\n\n\n

\u4ec0\u4e48\u7c7b\u578b\u7684\u7a0b\u5e8f\u9002\u5408\u5728GPU\u4e0a\u8fd0\u884c\uff1f<\/p>\n\n\n\n

\uff081\uff09\u8ba1\u7b97\u5bc6\u96c6\u578b\u7684\u7a0b\u5e8f\u3002\u6240\u8c13\u8ba1\u7b97\u5bc6\u96c6\u578b(Compute-intensive)\u7684\u7a0b\u5e8f\uff0c\u5c31\u662f\u5176\u5927\u90e8\u5206\u8fd0\u884c\u65f6\u95f4\u82b1\u5728\u4e86\u5bc4\u5b58\u5668\u8fd0\u7b97\u4e0a\uff0c\u5bc4\u5b58\u5668\u7684\u901f\u5ea6\u548c\u5904\u7406\u5668\u7684\u901f\u5ea6\u76f8\u5f53\uff0c\u4ece\u5bc4\u5b58\u5668\u8bfb\u5199\u6570\u636e\u51e0\u4e4e\u6ca1\u6709\u5ef6\u65f6\u3002\u53ef\u4ee5\u505a\u4e00\u4e0b\u5bf9\u6bd4\uff0c\u8bfb\u5185\u5b58\u7684\u5ef6\u8fdf\u5927\u6982\u662f\u51e0\u767e\u4e2a\u65f6\u949f\u5468\u671f\uff1b\u8bfb\u786c\u76d8\u7684\u901f\u5ea6\u5c31\u4e0d\u8bf4\u4e86\uff0c\u5373\u4fbf\u662fSSD, \u4e5f\u5b9e\u5728\u662f\u592a\u6162\u4e86\u3002<\/p>\n\n\n\n

\uff082\uff09\u6613\u4e8e\u5e76\u884c\u7684\u7a0b\u5e8f\u3002GPU\u5176\u5b9e\u662f\u4e00\u79cdSIMD(Single Instruction Multiple Data)\u67b6\u6784\uff0c \u4ed6\u6709\u6210\u767e\u4e0a\u5343\u4e2a\u6838\uff0c\u6bcf\u4e00\u4e2a\u6838\u5728\u540c\u4e00\u65f6\u95f4\u6700\u597d\u80fd\u505a\u540c\u6837\u7684\u4e8b\u60c5\u3002<\/p>\n\n\n\n

CPU\u4f1a\u5229\u7528\u8f83\u9ad8\u7684\u4e3b\u9891\u3001cache\u3001\u5206\u652f\u9884\u6d4b\u7b49\u6280\u672f\uff0c\u4f7f\u5904\u7406\u6bcf\u6761\u6307\u4ee4\u6240\u9700\u7684\u65f6\u95f4\u5c3d\u53ef\u80fd\u5c11\uff0c\u4ece\u800c\u964d\u4f4e\u5177\u6709\u590d\u6742\u8df3\u8f6c\u5206\u652f\u7a0b\u5e8f\u6267\u884c\u6240\u9700\u7684\u65f6\u95f4\u3002GPU\u5219\u901a\u8fc7\u6570\u91cf\u4e27\u5fc3\u75c5\u72c2\u7684\u6d41\u5904\u7406\u5668\u5b9e\u73b0\u5927\u91cf\u7ebf\u7a0b\u5e76\u884c\uff0c\u4f7f\u540c\u65f6\u8d70\u4e00\u6761\u6307\u4ee4\u7684\u6570\u636e\u53d8\u591a\uff0c\u4ece\u800c\u63d0\u9ad8\u6570\u636e\u7684\u541e\u5410\u91cf\u3002<\/p>\n\n\n\n

\u4e3e\u4e2aGPU\u901a\u7528\u8ba1\u7b97\u6559\u6750\u4e0a\u6bd4\u8f83\u5e38\u89c1\u7684\u4f8b\u5b50\uff0c\u4e00\u4e2a\u5411\u91cf\u76f8\u52a0\u7684\u7a0b\u5e8f\uff0c\u4f60\u53ef\u4ee5\u8ba9CPU\u8dd1\u4e00\u4e2a\u5faa\u73af\uff0c\u6bcf\u4e2a\u5faa\u73af\u5bf9\u4e00\u4e2a\u5206\u91cf\u505a\u52a0\u6cd5\uff0c\u4e5f\u53ef\u4ee5\u8ba9GPU\u540c\u65f6\u5f00\u5927\u91cf\u7ebf\u7a0b\uff0c\u6bcf\u4e2a\u5e76\u884c\u7684\u7ebf\u7a0b\u5bf9\u5e94\u4e00\u4e2a\u5206\u91cf\u7684\u76f8\u52a0\u3002CPU\u8dd1\u5faa\u73af\u7684\u65f6\u5019\u6bcf\u6761\u6307\u4ee4\u6240\u9700\u65f6\u95f4\u4e00\u822c\u4f4e\u4e8eGPU\uff0c\u4f46GPU\u56e0\u4e3a\u53ef\u4ee5\u5f00\u5927\u91cf\u7684\u7ebf\u7a0b\u5e76\u884c\u5730\u8dd1\uff0c\u5177\u6709SIMD\uff08\u51c6\u786e\u5730\u8bf4\u662fSIMT\uff09\u7684\u4f18\u52bf\u3002<\/p>\n\n\n\n

\u56e0\u6b64\u5bf9\u4e8e\u56fe\u5f62\u56fe\u50cf\u8ba1\u7b97\u3001\u5929\u6c14\u9884\u62a5\u4ee5\u53ca\u795e\u7ecf\u7f51\u7edc\u7b49\u90fd\u9002\u7528\u4e8eGPU\uff0c\u54e6\u5bf9\u4e86\uff0cGPU\u8fd8\u9002\u5408\u7528\u6765\u6316\u77ff\u3002<\/p>\n\n\n\n

\u4e3a\u4ec0\u4e48\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57dfGPU\u5341\u5206\u76db\u884c\uff1f<\/h5>\n\n\n\n

\u6df1\u5ea6\u5b66\u4e60\u662f\u6a21\u62df\u4eba\u8111\u795e\u7ecf\u7cfb\u7edf\u800c\u5efa\u7acb\u7684\u6570\u5b66\u7f51\u7edc\u6a21\u578b\uff0c\u8fd9\u4e2a\u6a21\u578b\u7684\u6700\u5927\u7279\u70b9\u662f\u9700\u8981\u5927\u6570\u636e\u6765\u8bad\u7ec3\u3002<\/mark><\/p>\n\n\n\n

\u56e0\u6b64\uff0c\u4eba\u5de5\u667a\u80fd\u9886\u57df\u5bf9\u8ba1\u7b97\u80fd\u529b\u7684\u8981\u6c42\u5c31\u662f\u9700\u8981\u5927\u91cf\u7684\u5e76\u884c\u7684\u91cd\u590d\u8ba1\u7b97\uff0cGPU\u6b63\u597d\u6709\u8fd9\u4e2a\u4e13\u957f\uff0c\u65f6\u52bf\u9020\u82f1\u96c4\uff0c\u56e0\u6b64GPU\u5c31\u51fa\u5c71\u62c5\u6b64\u91cd\u4efb\u3002<\/strong><\/p>\n\n\n\n

\u8bb2\u5230\u8fd9\u91cc\uff0c\u4e5f\u4e0d\u59a8\u63d0\u4e00\u4e0b\u6700\u8fd1\u51e0\u5e74\u5e02\u573a\u4e0a\u5f88\u706b\u7684AI\u82af\u7247\u3002\u8fd9\u4e2a\u5e02\u573a\u76ee\u524d\u4e3b\u6d41\u5206\u4e3aGPGPU\uff08GeneralPurposeGPU\uff09\u6d3e\u3001FPGA\u6d3e\u548cASIC\u6d3e\uff0c\u7279\u522b\u662f\u5728\u8bad\u7ec3\u65b9\u9762\u76ee\u524dGPGPU\u5360\u636e\u7edd\u5bf9\u4e3b\u5bfc\u5730\u4f4d\uff1a<\/p>\n\n\n\n

\u5728\u4f17\u591a\u7684GPU\u5382\u5546\u4e2d\uff0cNvidia\u4ee5\u5176\u4ea7\u54c1\u9ad8\u5ea6\u7684\u901a\u7528\u6027\uff08\u652f\u6301\u51e0\u4e4e\u6240\u6709\u4e3b\u6d41AI\u7b97\u6cd5\uff09\u3001\u6210\u719f\u7684\u8f6f\u4ef6\u751f\u6001\uff08CUDA\u51e0\u4e4e\u6210\u4e3a\u6df1\u5ea6\u5b66\u4e60\u7b97\u6cd5\u8bbe\u8ba1\u5fc5\u5907\u5de5\u5177\uff09\u53ca\u4e0d\u65ad\u8fed\u4ee3\u4f18\u5316\u7684\u4ea7\u54c1\u77e9\u9635\uff0c\u6210\u4e3a\u9876\u7ea7\u7684GPGPU\u4f9b\u5e94\u5546\uff1bGoogleTPU\u5728\u652f\u6301\u67d0\u4e9b\u56fa\u5316\u7684AI\u7b97\u6cd5\u65f6\u53ef\u8fbe\u5230\u6781\u9ad8\u7684\u6027\u80fd\/\u529f\u8017\u6bd4\u3002<\/p>\n\n\n\n

\u4ece\u957f\u671f\u6765\u770b\uff0cGPGPU\u7684\u9ad8\u5ea6\u901a\u7528\u6027\u548c\u53ef\u7f16\u7a0b\u6027\uff0c\u65e0\u7591\u5bf9\u4e8e\u9ad8\u901f\u8fed\u4ee3\u7684AI\u548c\u5176\u4ed6\u65b0\u578b\u7b97\u6cd5\u5177\u5907\u66f4\u597d\u7684\u9002\u5e94\u6027\u548c\u53d1\u5c55\u6f5c\u529b\uff0c\u52a0\u4e0a\u5176\u4ea7\u54c1\u7ecf\u8fc7\u957f\u671f\u7684\u8fed\u4ee3\u548c\u4f18\u5316\u8bbe\u8ba1\u66f4\u52a0\u6210\u719f\uff0c\u56e0\u6b64GPGPU\u6210\u4e3a\u8fd1\u5e74\u6765\u7b97\u6cd5\u516c\u53f8\u7684\u4e3b\u6d41\u786c\u4ef6\u52a0\u901f\u5de5\u5177\uff08\u7279\u522b\u662f\u89c4\u6a21\u8f83\u5927\u7684NN\u52a0\u901f\uff09\u3002<\/p>\n\n\n\n

\u5f53GPU\u9047\u5230\u4eba\u5de5\u667a\u80fd\u540e\uff0c\u5f00\u59cb\u5927\u653e\u5f02\u5f69\u30022006\u5e74\uff0c\u82f1\u4f1f\u8fbe\u63a8\u51fa\u57fa\u4e8eGPU\u7684CUDA\u5f00\u53d1\u5e73\u53f0\uff0c\u5f00\u53d1\u8005\u53ef\u4ee5\u901a\u8fc7\u8fd9\u4e00\u5e73\u53f0\uff0c\u4f7f\u7528C\u8bed\u8a00\u7f16\u5199\u7a0b\u5e8f\u4ee5\u89e3\u51b3\u590d\u6742\u7684\u8ba1\u7b97\u95ee\u9898\uff0c\u6362\u8a00\u4e4b\uff0c\u539f\u672c\u53ea\u7528\u505a3D\u6e32\u67d3\u7684GPU\u53d8\u5f97\u66f4\u52a0\u901a\u7528\uff0c\u53ef\u6267\u884c\u7684\u4efb\u52a1\u66f4\u52a0\u591a\u6837\u3002<\/p>\n\n\n\n

2009\u5e74\uff0c\u65af\u5766\u798f\u5927\u5b66\u7684Raina\u3001Madhavan\u53ca\u5434\u6069\u8fbe\u5728\u4e00\u7bc7\u8bba\u6587\u4e2d\u8bba\u8ff0\u4e86GPU\u5728\u6df1\u5ea6\u5b66\u4e60\u65b9\u9762\u76f8\u5bf9CPU\u7684\u5927\u5e45\u4f18\u52bf(http:\/\/robotics.stanford.edu\/~ang\/papers\/icml09-LargeScaleUnsupervisedDeepLearningGPU.pdf<\/a>)\uff0c\u5c06AI\u8bad\u7ec3\u65f6\u95f4\u4ece\u51e0\u5468\u7f29\u77ed\u81f3\u51e0\u5c0f\u65f6\u3002\u8fd9\u7bc7\u8bba\u6587\u4e3a\u4eba\u5de5\u667a\u80fd\u7684\u786c\u4ef6\u5b9e\u73b0\u6307\u660e\u4e86\u65b9\u5411\u3002GPU\u5927\u5927\u52a0\u901f\u4e86AI\u4ece\u8bba\u6587\u8d70\u5411\u73b0\u5b9e\u7684\u8fc7\u7a0b\u3002<\/p>\n\n\n\n

\u53c2\u8003\u94fe\u63a5\uff1a<\/h5>\n\n\n\n

CPU\u548cGPU\u7684\u4e0d\u540c
https:\/\/mp.weixin.qq.com\/s\/w_g35fttdhX1V6YUiujYDw<\/a><\/p>\n\n\n\n

CPU\u4e0eGPU\u5230\u5e95\u6709\u4ec0\u4e48\u533a\u522b\uff1f
https:\/\/mp.weixin.qq.com\/s\/fCuifWU-9tXMJLW_fLZRew<\/a><\/p>\n\n\n\n

\u7b80\u5355\u804a\u804aCPU\u548cGPU\u7684\u533a\u522b
https:\/\/mp.weixin.qq.com\/s\/dBn_Uiln4wacRsqpFXMuGg<\/a><\/p>\n\n\n\n

FPGA\u3001 CPU\u3001GPU\u3001ASIC\u7684\u533a\u522b
https:\/\/mp.weixin.qq.com\/s\/8d6QU4AVuXRJRk4l2KEdQg<\/a><\/p>\n\n\n\n

\u4ec0\u4e48\u662fGPU\uff1f\u8ddfCPU\u6709\u4ec0\u4e48\u533a\u522b\uff1f\u7ec8\u4e8e\u6709\u4eba\u8bb2\u660e\u767d\u4e86
https:\/\/mp.weixin.qq.com\/s\/MqgWdCuPTyJqFkJdLrB9pg<\/a><\/p>\n\n\n\n

\u7279\u65af\u62c9\u7684AI\u91ce\u5fc3\uff1a\u5411\u4eba\u7c7b\u9884\u8b66\uff0c\u7ed9\u7845\u57fa\u5e26\u8def
https:\/\/mp.weixin.qq.com\/s\/zV1HUgIQTAVkxmZDbWKPPw<\/a><\/p>\n\n\n\n

=END=<\/p>\n","protected":false},"excerpt":{"rendered":"

=Start= \u7f18\u7531\uff1a \u81ea\u4ece chatGPT \u6a2a\u7a7a\u51fa\u4e16\u4e4b\u540e\uff0c\u5173\u4e8e AI \u7684\u8ba8\u8bba\u5c42\u51fa\u4e0d\u7a77\uff0c\u4e3a\u4e86\u907f\u514d\u4fe1\u606f\u8131\u8282\uff0c\u9700\u8981 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[23,1275],"tags":[1931,946,1928,1929,1930,1482,1478],"class_list":["post-5423","post","type-post","status-publish","format-standard","hentry","category-knowledgebase-2","category-1275","tag-chatgpt","tag-cpu","tag-gpu","tag-nvidia","tag-simd","tag-1482","tag-1478"],"views":3245,"_links":{"self":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/5423","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/comments?post=5423"}],"version-history":[{"count":1,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/5423\/revisions"}],"predecessor-version":[{"id":5424,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/posts\/5423\/revisions\/5424"}],"wp:attachment":[{"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/media?parent=5423"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/categories?post=5423"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ixyzero.com\/blog\/wp-json\/wp\/v2\/tags?post=5423"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}