{"id":199647,"date":"2020-11-04T12:41:18","date_gmt":"2020-11-04T04:41:18","guid":{"rendered":"http:\/\/4563.org\/?p=199647"},"modified":"2020-11-04T12:41:18","modified_gmt":"2020-11-04T04:41:18","slug":"tensorflow-2-%e7%8e%b0%e5%b7%b2%e6%94%af%e6%8c%81%e4%bd%bf%e7%94%a8-macos-%e4%b8%8a%e5%8f%af%e7%94%a8%e7%9a%84-gpu-%e8%bf%9b%e8%a1%8c%e6%a8%a1%e5%9e%8b%e5%8a%a0%e9%80%9f%ef%bc%88%e5%8c%85%e6%8b%ac-app","status":"publish","type":"post","link":"http:\/\/4563.org\/?p=199647","title":{"rendered":"TensorFlow 2 \u73b0\u5df2\u652f\u6301\u4f7f\u7528 macOS \u4e0a\u53ef\u7528\u7684 GPU \u8fdb\u884c\u6a21\u578b\u52a0\u901f\uff08\u5305\u62ec Apple Silicon M1\uff09"},"content":{"rendered":"
\n
\n
\n

TensorFlow 2 \u73b0\u5df2\u652f\u6301\u4f7f\u7528 macOS \u4e0a\u53ef\u7528\u7684 GPU \u8fdb\u884c\u6a21\u578b\u52a0\u901f\uff08\u5305\u62ec Apple Silicon M1\uff09 <\/h1>\n

<\/p>\n

\n
\u8cc7\u6df1\u5927\u4f6c : mmtromsb456 <\/span> <\/i> 9<\/span> <\/div>\n
<\/div>\n<\/p><\/div>\n<\/p><\/div>\n<\/p><\/div>\n
<\/p>\n

\u8be6\u60c5\u53ef\u89c1 https:\/\/blog.tensorflow.org\/2020\/11\/accelerating-tensorflow-performance-on-mac.html?m=1<\/p>\n<\/p><\/div>\n

\u5927\u4f6c\u6709\u8a71\u8aaa<\/b> (19<\/span>) <\/div>\n
<\/div>\n<\/p><\/div>\n<\/p><\/div>\n
    \n
  • \n
    \n
    \n
    \u4e3b<\/span> \u8cc7\u6df1\u5927\u4f6c : mmtromsb456 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u7b80\u5355\u6765\u8bf4 apple fork \u4e86\u4e00\u4efd TensorFlow \u5e76\u5c06 ML Compute \u6846\u67b6\u7528\u4f5c\u5176\u4e2d\uff0cM1 \u6027\u80fd\u5927\u6982\u76f8\u5f53\u4e8e 1080 \uff5e 1080Ti \u4e4b\u95f4 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : ltltfuture <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u63a5\u4e0b\u6765\u5c31\u770b pytorch \u5982\u4f55\u4e86 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : xfrgux <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    @mmtromsb456 \u8bf7\u95ee\u8fd9\u4e2a\u76f8\u5f53\u4e8e 1080 \uff5e 1080Ti \u6709\u6570\u636e\u6765\u6e90\u5417 \u600e\u4e48\u6bd4\u7684\u5462 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u4e3b<\/span> \u8cc7\u6df1\u5927\u4f6c : mmtromsb456 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    @xfrgux https:\/\/twitter.com\/spurpura\/status\/1329168059647488000?s=21
    \u6d4b\u8bd5\u7684\u6570\u636e\u6a21\u578b\u662f ResNet50v1.5,v2 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : zhusimaji <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    apple \u4e0d\u8bb2\u6b66\u5fb7\u554a\uff01 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : tzm41 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u723d\uff0c\u4eca\u665a\u8bd5\u8bd5 resnet \u3002 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : uxstone <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    Apple \u6740\u75af\u4e86 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : lollxxox <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u5e0c\u671b I \u5bb6 A \u5bb6\u548c N \u5bb6\u8017\u5b50\u5c3e\u6c41\uff0c\u4e0d\u8981\u518d\u6324\u7259\u818f……. <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : chich14 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u70bc\u4e39\uff0c\u98ce\u6247\u8f6c\u4e86\u5417\uff1f <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : CommandZi <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u4e13\u7528\u82af\u7247\uff0c\u4e0d\u80fd\u53cd\u6620\u771f\u5b9e\u6027\u80fd\uff0c\u4e0b\u4e00\u4e2a <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : agagega <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u80fd\u652f\u6301 NPU \u5417\uff1f\u5229\u7528\u8d77\u6765\u611f\u89c9\u4f1a\u66f4\u731b\u554a <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : zhaidoudou123 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u7528\u81ea\u5df1\u673a\u5668\u90fd\u662f\u73a9\u7968\u6027\u8d28\uff0c\u73b0\u5728 m1 \u7684\u73a9\u7968\u6027\u80fd\u8d77\u6765\u4e86\uff0c\u5bf9\u4e8e\u521d\u5b66\u8005\u6765\u8bf4 n \u5361\u7b14\u8bb0\u672c\u4f3c\u4e4e\u5df2\u7ecf\u4e0d\u662f\u5fc5\u987b\u4e86 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : nnqijiu <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    TensorFlow \u4e0d\u662f\u5df2\u7ecf\u653e\u5f03 mac \u652f\u6301\u4e86\u5417\uff1f <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : coolmenu <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u4ee5\u540e\u80fd\u4e0d\u80fd\u5927\u91cf\u8d2d\u4e70 M1 \u82af\u7247\uff0c\u7ec4\u4e00\u4e2a\u5e76\u53d1\u7f51\u7edc\u300210w \u7684 cpu\uff0c\u6bd4 1080ti \u53ef\u662f\u7701\u7535\u591a\u5566\uff01\uff01 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : lqf96 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    @ltltfuture \u597d\u5e0c\u671b\u65e9\u65e5\u652f\u6301 pytorch\uff0c\u4e0d\u8fc7 pytorch \u4f3c\u4e4e\u4e5f\u5df2\u7ecf\u5728\u7528 Metal Performance Shaders \u52a0\u901f\u8bad\u7ec3\u548c\u8fd0\u884c\u4e86\uff08 https:\/\/github.com\/pytorch\/pytorch\/pull\/46112 \uff09…
    @agagega \u76ee\u524d ML Compute \u53ea\u652f\u6301 CPU \u548c GPU\uff0c\u4f46\u662f ML Compute \u91cc\u9762\u6709 ANE \u76f8\u5173\u7684 symbol\uff0c\u6240\u4ee5\u4e0d\u6392\u9664\u662f\u76ee\u524d Neural Engine \u7684\u52a0\u901f\u4e0d\u6210\u719f\uff08\u56e0\u4e3a MLC \u662f macOS 11 \u548c iOS 14 \u521a\u52a0\u8fdb\u6765\u7684\uff09\uff0c\u4ee5\u540e\u53ef\u80fd\u5c31\u4f1a\u652f\u6301…<\/p>\n

    “`
    00007fff2a15aae2 t +[MLCDeviceANE hasANE]
    00007fff2a22b934 t +[MLCDeviceANE hasANE].cold.1
    00007fff2a15b672 t -[MLCDeviceANE .cxx_destruct]
    00007fff2a15ad57 t -[MLCDeviceANE allocateDeviceMemoryForTensor:]
    00007fff2a22bcad t -[MLCDeviceANE allocateDeviceMemoryForTensor:].cold.1
    00007fff2a15b05c t -[MLCDeviceANE broadcastTensor:]
    00007fff2a22c124 t -[MLCDeviceANE broadcastTensor:].cold.1
    00007fff2a15b0b1 t -[MLCDeviceANE broadcastTensor:batchSize:]
    00007fff2a22c1a3 t -[MLCDeviceANE broadcastTensor:batchSize:].cold.1
    00007fff2a15b309 t -[MLCDeviceANE commitAndWaitForCompletion:]
    00007fff2a22c51c t -[MLCDeviceANE commitAndWaitForCompletion:].cold.1
    00007fff2a15b35e t -[MLCDeviceANE commitAndWaitForCompletion:enableProfiling:graphExecutionTime:graphResultTensor:]
    00007fff2a22c59b t -[MLCDeviceANE commitAndWaitForCompletion:enableProfiling:graphExecutionTime:graphResultTensor:].cold.1<\/p>\n

    …
    “` <\/div>\n<\/p><\/div>\n<\/li>\n

  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : sharpy <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u6709\u95ee\u9898\uff0c\u6211\u5b89\u88c5\u4e86 mac \u63d0\u4f9b\u7684 tf \u5305\uff0c\u8fd0\u884c mnist demo \u65f6\uff0c\u63d0\u793a\u5728 eager \u6a21\u5f0f\u4e0b gpu \u901f\u5ea6\u6162\u3002
    \u7528 cpu \u65f6\u8017\u65f6 2s\uff0c\u7528 gpu \u65f6\u8017\u65f6 9s \u3002 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : wwqgtxx <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    @sharpy \u505a\u6027\u80fd\u6d4b\u8bd5\u672c\u6765\u5c31\u4e0d\u8be5\u7528 eager \u6a21\u5f0f\uff0c\u800c\u4e14\u4e00\u822c\u8981\u591a\u8fd0\u884c\u51e0\u4e2a\u5faa\u73af\u6765\u53bb\u6389 gpu \u521d\u59cb\u5316\u7684\u65f6\u95f4 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : tzm41 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    \u6d4b\u4e86\u4e00\u4e0b\u7b80\u5355\u7684 CNN\uff0c\u4e0d\u77e5\u4e3a\u4f55\u5728\u6211\u7684 MBP 16 \u4e0a batch \u4e4b\u95f4\u4f1a\u5361\u4f4f\u3002 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n
    \n
    \n
    \u8cc7\u6df1\u5927\u4f6c : volvo007 <\/span> <\/div>\n
    <\/i> <\/span> <\/div>\n<\/p><\/div>\n
    Metal \u7684\u8bdd\uff0c\u80fd\u5916\u63a5 a \u5361\u63d0\u901f\u5417 <\/div>\n<\/p><\/div>\n<\/li>\n
  • \n","protected":false},"excerpt":{"rendered":"

    TensorFlow 2 \u73b0\u5df2\u652f\u6301…<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[],"tags":[],"_links":{"self":[{"href":"http:\/\/4563.org\/index.php?rest_route=\/wp\/v2\/posts\/199647"}],"collection":[{"href":"http:\/\/4563.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/4563.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/4563.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/4563.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=199647"}],"version-history":[{"count":0,"href":"http:\/\/4563.org\/index.php?rest_route=\/wp\/v2\/posts\/199647\/revisions"}],"wp:attachment":[{"href":"http:\/\/4563.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=199647"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/4563.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=199647"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/4563.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=199647"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}