{"id":8516,"date":"2026-06-09T14:12:55","date_gmt":"2026-06-09T06:12:55","guid":{"rendered":"https:\/\/ab123.xyz\/?p=8516"},"modified":"2026-06-09T14:12:55","modified_gmt":"2026-06-09T06:12:55","slug":"apple-m4-ultra-neural-engine-%e4%b8%8e-coreml-%e6%b7%b1%e5%ba%a6%e9%9b%86%e6%88%90%ef%bc%9aai-%e5%bc%80%e5%8f%91%e8%80%85%e7%9a%84%e6%96%b0%e5%88%a9%e5%99%a8","status":"publish","type":"post","link":"https:\/\/ab123.xyz\/?p=8516","title":{"rendered":"Apple M4 Ultra Neural Engine \u4e0e CoreML \u6df1\u5ea6\u96c6\u6210\uff1aAI \u5f00\u53d1\u8005\u7684\u65b0\u5229\u5668"},"content":{"rendered":"

\u82f9\u679c\u6700\u65b0\u53d1\u5e03\u7684 M4 Ultra \u82af\u7247\uff0c\u51ed\u501f\u5176\u9769\u547d\u6027\u7684\u795e\u7ecf\u7f51\u7edc\u5f15\u64ce\uff08Neural Engine\uff09\u548c\u4e0e CoreML \u6846\u67b6\u7684\u6df1\u5ea6\u878d\u5408\uff0c\u6b63\u5728\u91cd\u65b0\u5b9a\u4e49\u7aef\u4fa7\u4eba\u5de5\u667a\u80fd\u7684\u6027\u80fd\u8fb9\u754c\u3002\u636e WWDC 2025 \u516c\u5e03\u7684\u6280\u672f\u7ec6\u8282\uff0cM4 Ultra \u7684 Neural Engine \u62e5\u6709 256 \u6838\u8bbe\u8ba1\uff0c\u7b97\u529b\u8fbe\u5230 128 TOPS\uff0c\u76f8\u6bd4 M3 Ultra \u63d0\u5347\u8fd1\u4e24\u500d\u3002\u8fd9\u610f\u5473\u7740\u5f00\u53d1\u8005\u53ef\u4ee5\u5728 Mac Studio \u6216 Mac Pro \u4e0a\u76f4\u63a5\u8fd0\u884c\u5927\u578b\u8bed\u8a00\u6a21\u578b\uff08LLM\uff09\u3001\u56fe\u50cf\u751f\u6210\u6a21\u578b\u548c\u5b9e\u65f6\u89c6\u9891\u5206\u6790\u4efb\u52a1\u800c\u65e0\u9700\u4f9d\u8d56\u4e91\u7aef\u3002\u8bbf\u95ee \u5b98\u65b9\u7f51\u7ad9<\/a> \u53ef\u83b7\u53d6\u6700\u65b0 SDK \u548c\u793a\u4f8b\u4ee3\u7801\u3002<\/p>\n

\u6838\u5fc3\u529f\u80fd\u4e0e\u67b6\u6784\u5347\u7ea7<\/h2>\n

M4 Ultra \u7684 Neural Engine \u5e76\u975e\u7b80\u5355\u7684\u786c\u4ef6\u5806\u53e0\uff0c\u800c\u662f\u4e0e CoreML \u6846\u67b6\u8fdb\u884c\u4e86\u5168\u6808\u534f\u540c\u4f18\u5316\u3002\u901a\u8fc7\u7edf\u4e00\u5185\u5b58\u67b6\u6784\uff08UMA\uff09\uff0cCPU\u3001GPU \u548c Neural Engine \u53ef\u4ee5\u5171\u4eab\u9ad8\u8fbe 512GB \u7684\u7edf\u4e00\u5185\u5b58\uff0c\u6d88\u9664\u6570\u636e\u642c\u8fd0\u5ef6\u8fdf\u3002CoreML 4.0 \u7248\u672c\u65b0\u589e\u4e86\u52a8\u6001\u56fe\u7f16\u8bd1\u529f\u80fd\uff0c\u80fd\u591f\u81ea\u52a8\u5c06 ONNX\u3001PyTorch \u7b49\u6a21\u578b\u8f6c\u6362\u4e3a\u9488\u5bf9 M4 Ultra \u7684\u91cf\u5316\u6307\u4ee4\uff0c\u8ba9\u5f00\u53d1\u8005\u65e0\u9700\u624b\u52a8\u8c03\u4f18\u5373\u53ef\u83b7\u5f97 3-5 \u500d\u63a8\u7406\u901f\u5ea6\u63d0\u5347\u3002<\/p>\n

\u5173\u952e\u4f18\u52bf\u4e00\u89c8<\/h3>\n
    \n
  • \u4f4e\u529f\u8017\u9ad8\u6027\u80fd\uff1a<\/strong>\u6bcf TOPS \u529f\u8017\u4ec5\u4e3a 0.8W\uff0c\u9002\u5408\u957f\u65f6\u95f4 AI \u63a8\u7406\u4efb\u52a1\u3002<\/li>\n
  • \u9690\u79c1\u5b89\u5168\uff1a<\/strong>\u6240\u6709\u6570\u636e\u5728\u8bbe\u5907\u672c\u5730\u5904\u7406\uff0c\u65e0\u9700\u4e0a\u4f20\u4e91\u7aef\uff0c\u7b26\u5408 GDPR \u548c\u300a\u4e2a\u4eba\u4fe1\u606f\u4fdd\u62a4\u6cd5\u300b\u8981\u6c42\u3002<\/li>\n
  • \u751f\u6001\u517c\u5bb9\uff1a<\/strong>\u652f\u6301 Hugging Face \u6a21\u578b\u76f4\u63a5\u5bfc\u5165\uff0c\u8986\u76d6 Stable Diffusion\u3001Whisper\u3001YOLO \u7b49\u4e3b\u6d41\u6a21\u578b\u3002<\/li>\n<\/ul>\n

    \u5e94\u7528\u573a\u666f\u5168\u8986\u76d6<\/h2>\n

    M4 Ultra + CoreML \u7ec4\u5408\u5df2\u843d\u5730\u591a\u4e2a\u9886\u57df\uff1a\u5728\u533b\u7597\u5f71\u50cf\u5206\u6790\u4e2d\uff0c\u653e\u5c04\u79d1\u533b\u751f\u53ef\u5728 0.2 \u79d2\u5185\u5b8c\u6210 X \u5149\u7247\u75c5\u7076\u6807\u8bc6\uff1b\u5728\u521b\u610f\u8bbe\u8ba1\u9886\u57df\uff0c\u8bbe\u8ba1\u5e08\u5bf9 4K \u89c6\u9891\u8fdb\u884c\u5b9e\u65f6\u98ce\u683c\u8fc1\u79fb\uff0c\u5ef6\u8fdf\u4f4e\u4e8e 10 \u6beb\u79d2\uff1b\u5728\u81ea\u52a8\u9a7e\u9a76\u6a21\u62df\u4e2d\uff0c\u5de5\u7a0b\u5e08\u4f7f\u7528 Neural Engine \u52a0\u901f\u4f20\u611f\u5668\u878d\u5408\u7b97\u6cd5\uff0c\u8bad\u7ec3\u6548\u7387\u63d0\u5347 4 \u500d\u3002\u82f9\u679c\u8fd8\u63a8\u51fa\u4e86 CoreML Benchmark Suite<\/strong>\uff0c\u5e2e\u52a9\u5f00\u53d1\u8005\u5bf9\u6bd4\u4e0d\u540c\u82af\u7247\u4e0a\u7684\u63a8\u7406\u6027\u80fd\u3002<\/p>\n

    \u5982\u4f55\u4f7f\u7528\u4e0e\u5f00\u53d1\u5efa\u8bae<\/h3>\n

    \u5f00\u53d1\u8005\u53ea\u9700\u5728 Xcode 16 \u4e2d\u5f15\u5165 CoreML \u5e93\uff0c\u8c03\u7528 MLModelConfiguration.computeUnits = .all<\/code> \u5373\u53ef\u542f\u7528\u5168\u90e8 Neural Engine \u6838\u5fc3\u3002\u5bf9\u4e8e\u81ea\u5b9a\u4e49\u6a21\u578b\uff0c\u53ef\u4f7f\u7528 CoreML Tools<\/strong> \u8fdb\u884c\u8f6c\u6362\u5e76\u5f00\u542f\u201c\u6df7\u5408\u7cbe\u5ea6\u201d\u4f18\u5316\u3002\u82f9\u679c\u5b98\u65b9\u5efa\u8bae\u642d\u914d Create ML 4.0<\/strong> \u8fdb\u884c\u4f4e\u4ee3\u7801\u8bad\u7ec3\uff0c\u6216\u76f4\u63a5\u4f7f\u7528 MLX Framework<\/strong> \u5728 M4 Ultra \u4e0a\u5fae\u8c03\u5927\u6a21\u578b\u3002<\/p>\n

    \u6700\u65b0\u65b0\u95fb\uff1aM4 Ultra \u52a9\u529b AI \u5b9e\u65f6\u7ffb\u8bd1\u8fdb\u5165\u65b0\u7eaa\u5143<\/h2>\n

    \u636e TechCrunch<\/em> \u62a5\u9053\uff0c\u5728\u521a\u7ed3\u675f\u7684 WWDC 2025 \u4e3b\u9898\u6f14\u8bb2\u4e2d\uff0c\u82f9\u679c\u6f14\u793a\u4e86\u57fa\u4e8e M4 Ultra \u7684\u5b9e\u65f6\u8bed\u97f3\u7ffb\u8bd1\u7cfb\u7edf\uff0c\u53ef\u5c06 100 \u79cd\u8bed\u8a00\u5728 300 \u6beb\u79d2\u5185\u5b8c\u6210\u8bc6\u522b\u3001\u7ffb\u8bd1\u548c\u8bed\u97f3\u5408\u6210\uff0c\u540c\u65f6\u5904\u7406 8 \u8def\u97f3\u9891\u6d41\u3002\u8be5\u6280\u672f\u5df2\u96c6\u6210\u5230 macOS 15 \u548c iOS 19 \u7cfb\u7edf\u4e2d\uff0c\u5f00\u53d1\u8005\u53ef\u901a\u8fc7 CoreML Speech API \u8c03\u7528\u3002\u8fd9\u4e00\u7a81\u7834\u6027\u8fdb\u5c55\u8ba9\u8de8\u56fd\u4f1a\u8bae\u3001\u5728\u7ebf\u6559\u80b2\u7b49\u573a\u666f\u7684\u6c9f\u901a\u969c\u788d\u5927\u5e45\u964d\u4f4e\u3002<\/p>\n

    \u968f\u7740 M4 Ultra \u7684\u91cf\u4ea7\uff0c\u82f9\u679c\u8fdb\u4e00\u6b65\u5de9\u56fa\u4e86\u5176\u5728\u7aef\u4fa7 AI \u8ba1\u7b97\u7684\u9886\u5bfc\u5730\u4f4d\u3002\u5bf9\u4e8e\u4f01\u4e1a\u7ea7 AI \u5e94\u7528\u5f00\u53d1\uff0c\u5efa\u8bae\u4f18\u5148\u9009\u62e9\u642d\u8f7d M4 Ultra \u7684 Mac Studio\uff0c\u5e76\u7ed3\u5408 CoreML \u7684\u5bb9\u5668\u5316\u90e8\u7f72\u65b9\u6848\u5b9e\u73b0\u5feb\u901f\u4e0a\u7ebf\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    \u82f9\u679c\u6700\u65b0\u53d1\u5e03\u7684 M4 Ultra \u82af\u7247\uff0c\u51ed\u501f\u5176\u9769\u547d\u6027\u7684\u795e\u7ecf\u7f51\u7edc\u5f15\u64ce\uff08Neural Engine\uff09\u548c\u4e0e Core […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[4],"tags":[10320,10613,10614,10635,10636],"class_list":["post-8516","post","type-post","status-publish","format-standard","hentry","category-4","tag-ai","tag-apple-m4-ultra","tag-coreml","tag-neural-engine","tag-10636"],"_links":{"self":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/8516","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=8516"}],"version-history":[{"count":1,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/8516\/revisions"}],"predecessor-version":[{"id":8517,"href":"https:\/\/ab123.xyz\/index.php?rest_route=\/wp\/v2\/posts\/8516\/revisions\/8517"}],"wp:attachment":[{"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=8516"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=8516"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ab123.xyz\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=8516"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}