{"id":6,"date":"2025-12-10T09:01:30","date_gmt":"2025-12-10T00:01:30","guid":{"rendered":"https:\/\/www.em99.tech\/?p=6"},"modified":"2025-12-11T16:39:32","modified_gmt":"2025-12-11T07:39:32","slug":"gpusoroban%e3%81%95%e3%82%93%e3%82%92%e4%bd%bf%e3%81%a3%e3%81%a6%e3%81%bf%e3%81%9f1","status":"publish","type":"post","link":"https:\/\/www.em99.tech\/?p=6","title":{"rendered":"GPUSOROBAN\u3055\u3093\u3092\u4f7f\u3063\u3066\u307f\u305f(1)"},"content":{"rendered":"\n

\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001
GPUSOROBAN \u3067 GPU \u304c\u4f7f\u3048\u308b PyTorch \u74b0\u5883\u3092\u6700\u77ed\u3067\u69cb\u7bc9\u3059\u308b\u307e\u3067\u306e\u5b9f\u8df5\u30ed\u30b0<\/strong>
\u3092\u66f8\u3044\u3066\u3044\u307e\u3059\u3002<\/p>\n\n\n\n

\u77e5\u308a\u5408\u3044\u304cYoutube\u306b\u6295\u7a3f\u3057\u3066\u3044\u305f\u308a\u3001\u81ea\u5206\u3067\u3082\u3084\u3063\u3066\u307f\u305f\u3044\u304b\u306a\u3001\u3068\u601d\u3046\u3053\u3068\u3082\u3042\u3063\u3066\u3001\u4f8b\u3048\u3070\u6620\u308a\u8fbc\u3093\u3060\u4eba\u306e\u9854\u306b\u30e2\u30b6\u30a4\u30af\u5165\u308c\u305f\u308a\u3068\u304b\u3001\u3059\u3067\u306bAI\u3092\u5229\u7528\u3057\u305f\u30b5\u30fc\u30d3\u30b9\u3084\u30a2\u30d7\u30ea\u306a\u3093\u304b\u3082\u304d\u3063\u3068\u3042\u308b\u306f\u305a\u306a\u306e\u3067\u3059\u304c\u3001\u3042\u308b\u7a0b\u5ea6\u30ab\u30b9\u30bf\u30de\u30a4\u30ba\u3057\u305f\u308a\u3001\u3044\u308d\u3044\u308d\u8a66\u3057\u305f\u308a\u3082\u3057\u305f\u3044\u3058\u3083\u306a\u3044\u3067\u3059\u304b\u3002

\u81ea\u5206\u306e\u624b\u5143\u306e\u74b0\u5883\u306f\u4f7f\u3048\u305d\u3046\u306aGPU\u74b0\u5883\u304c\u306a\u3044\u3001\u3068\u3044\u3046\u3053\u3068\u3067\u3001\u4e00\u756a\u5b89\u4fa1\u306a\u30d7\u30e9\u30f3\u3060\u3068\u5f93\u91cf\u5236\u3067\u6642\u959350\u5186\u3068\u3044\u3046GPUSORABAN\u3055\u3093\u3092\u5229\u7528\u3059\u308b\u3053\u3068\u306b\u3057\u3066\u307f\u307e\u3057\u305f\u3002

\u3044\u3064\u3067\u3082\u6b62\u3081\u3089\u308c\u308b\u306e\u3067\u4ed5\u69d8\u3084\u3089\u4f55\u3084\u3089\u306f\u307b\u307c\u8abf\u3079\u305a…\u7b11\uff1c\u30a2\u30db\u3084

\u6700\u521ddocker\u3092\u4f7f\u3048\u308b\u30d7\u30e9\u30f3\u3092\u9078\u3093\u3060\u306e\u3067\u3059\u3002\u306a\u3093\u3060\u304b\u97ff\u304d\u3060\u3051\u3067\uff1c\u3084\u3063\u3071\u308a\u30a2\u30db
\u3067\u3001\u9069\u5f53\u306b\u8a66\u3057\u3066\u3044\u305f\u306e\u3067\u3059\u304c\u809d\u5fc3\u306eGPU\u304c\u4f7f\u3048\u306a\u3044…
\u7121\u8ad6docker\u3060\u304b\u3089\u3068\u3044\u3046\u30ef\u30b1\u3067\u306f\u3042\u308a\u307e\u305b\u3093\u3002\u79c1\u304c\u30c0\u30e1\u306a\u5974\u3060\u3063\u305f\u3068\u3044\u3046\u3060\u3051\u3067\u3002

\u305f\u3060\u305d\u306e\u6642\u306f\u539f\u56e0\u306e\u5207\u308a\u5206\u3051\u3082\u5c11\u3057\u3057\u305f\u304f\u3066\u3001s16-1-a-standard-ubs22-i\u3000\u3068\u3044\u3046\u30bf\u30a4\u30d7\u3092\u9078\u3073\u76f4\u3057\u307e\u3057\u305f\u3002\u30b9\u30da\u30c3\u30af\u30fb\u30d7\u30ea\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u60c5\u5831\u306f\u3053\u3061\u3089<\/a>\u3092\u53c2\u7167\u304f\u3060\u3055\u3044\u3002<\/p>\n\n\n\n

1.\u307e\u305a\u4f55\u306b\u30cf\u30de\u3063\u305f\u304b\uff08Torch\u3068CUDA\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u554f\u984c\uff09
\u30d7\u30ed\u30b0\u30e9\u30de\u30fc\u3067\u3082\u306a\u3044\u306e\u3067(\u8a00\u3044\u8a33)\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u306fpip\u3060\u308d\u3045\u3001\u3068\u306a\u3093\u3067\u3082pip\u3067\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u3066\u3044\u305f\u308f\u3051\u3067\u3059\u304c\u3001\u95a2\u9023\u30e2\u30b8\u30e5\u30fc\u30eb\u306btorch\u95a2\u9023\u304c\u3042\u308b\u3068\u3001\u52dd\u624b\u306b\u30d0\u30fc\u30b8\u30e7\u30f3\u30a2\u30c3\u30d7\u3057\u3066\u3057\u307e\u3046\u30ef\u30b1\u3067\u3059\u3002
\u3067\u3001\u30d0\u30fc\u30b8\u30e7\u30f3\u30a2\u30c3\u30d7\u3057\u306a\u3044\u3088\u3046\u306b\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u9032\u3081\u3066\u3001\u5fc5\u8981\u306a\u30e2\u30b8\u30e5\u30fc\u30eb\u30921\u500b\u305a\u3064\u305a\u3064\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u3066\u3044\u3063\u305f\u30ef\u30b1\u3067\u3059\u304c…torch\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u304c\u30a2\u30d7\u30ea\u304c\u8981\u6c42\u3059\u308b\u30d0\u30fc\u30b8\u30e7\u30f3\u3088\u308a\u53e4\u3044\u3001\u3068\u304b\u3067GPU\u3092\u4f7f\u3048\u306a\u3044\u3001\u30a2\u30d7\u30ea\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u3092\u843d\u3068\u3057\u3066\u3082\u4e0d\u8db3\u304c\u3042\u3063\u3066\u4f7f\u3048\u306a\u3044\u3001\u3068\u3044\u3063\u305f\u611f\u3058\u3067\u3084\u306f\u308aGPU\u3092\u4f7f\u3048\u306a\u3044\u6642\u9593\u304c\u904e\u304e\u3066\u3044\u304d…\u3002
GPUSORABAN\u3055\u3093\u306e\u8a18\u4e8b\u3092\u8aad\u307f\u8fd4\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n

\u30bc\u30ed\u304b\u3089\u5206\u304b\u308b\uff01PyTorch\u306eGPU\u8a2d\u5b9a\u3068\u5b9f\u8df5\u30ec\u30b7\u30d4<\/a><\/p>\n\n\n\n

\u201d\u5229\u7528\u3057\u3088\u3046\u3068\u3057\u3066\u3044\u308bPyTorch\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u304c\u3001\u3069\u306e\u30d0\u30fc\u30b8\u30e7\u30f3\u306eCUDA\u306b\u5bfe\u5fdc\u3057\u3066\u3044\u308b\u304b\u3092\u4e8b\u524d\u306b\u78ba\u8a8d\u3057\u307e\u3059\u3002\u201d<\/p>\n\n\n\n

\u307e\u3001\u307e\u3042\u3001\u305d\u3046\u3067\u3059\u3088\u306d\u3002\u666e\u901a\u306e\u3053\u3068\u3067\u3059\u3088\u306d\u6c57
\u3068\u3044\u3046\u3053\u3068\u3067OS\u3068\u30d0\u30fc\u30b8\u30e7\u30f3\u306b\u5408\u308f\u305b\u305f\u30b3\u30de\u30f3\u30c9\u306e\u8a18\u8f09\u304c\u3042\u308b\u516c\u5f0f\u30da\u30fc\u30b8<\/a>\u304b\u3089\u74b0\u5883\u306b\u3042\u3063\u305f\u30b3\u30de\u30f3\u30c9\u3092\u63a2\u3057\u3066\u5b9f\u884c\u3057\u3066\u307f\u307e\u3057\u305f\u3002<\/p>\n\n\n\n

2.\u3067\u3001\u5b9f\u969b\u306a\u306b\u3084\u3063\u305f\u304b<\/p>\n\n\n\n

\u30fbGPUSORABAN\u3055\u3093\u306e\u30a4\u30f3\u30b9\u30bf\u30f3\u30b9\u4f5c\u6210\u65b9\u6cd5\u306f\u304a\u305d\u3089\u304f\u52d8\u3067\u3082\u51fa\u6765\u308b\u306f\u305a…<\/p>\n\n\n\n

\u30fbMiniconda \u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb
\u3042\u308c\u3001wget\u3082\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u305f\u304b\u3082…<\/p>\n\n\n\n

$ wget https:\/\/repo.anaconda.com\/miniconda\/Miniconda3-latest-Linux-x86_64.sh
$ bash Miniconda3-latest-Linux-x86_64.sh<\/p>\n\n\n\n

\u30fbGPU \u3092\u691c\u8a3c\u3059\u308b\u305f\u3081\u306e\u74b0\u5883\u3092\u4f5c\u308b
conda\u3067\u4eee\u60f3\uff1f\u74b0\u5883\u307f\u305f\u3044\u306a\u3082\u306e\u3092\u4f5c\u3063\u3066\u4ed6\u306b\u5f71\u97ff\u3092\u4e0e\u3048\u306a\u3044\u3088\u3046\u306b\u3002
(Denoiser\u3092\u8a66\u3057\u305f\u304b\u3063\u305f\u306e\u3067\u540d\u524d\u306fdenoiserenv\u306b\u3057\u3066\u307f\u307e\u3057\u305f\u3002)

\u3067\u3001\u305d\u306e\u307e\u307e\u52e2\u3044\u3067conda\u3092\u4f7f\u7528\u3057\u3088\u3046\u3068\u3059\u308b\u3068\u6012\u3089\u308c\u308b\u306e\u3067\u8a8d\u8a3c\uff1f\u30b3\u30de\u30f3\u30c9\u3092<\/p>\n\n\n\n

$ conda tos accept –override-channels –channel https:\/\/repo.anaconda.com\/pkgs\/main
$ conda tos accept –override-channels –channel https:\/\/repo.anaconda.com\/pkgs\/r<\/p>\n\n\n\n

\u30fb\u4eee\u60f3\u74b0\u5883\u306e\u4f5c\u6210\u3068\u4eee\u60f3\u74b0\u5883\u5185\u3078\u5165\u308b<\/p>\n\n\n\n

$ conda create -n denoiserenv python=3.10 -y
$ conda activate denoiserenv<\/p>\n\n\n\n

python\u3082\u53e4\u3081\u306a\u306e\u306f\u524d\u6bb5\u306b\u66f8\u3044\u305f\u8272\u3005\u5931\u6557\u3057\u305f\u7d50\u679c\u3067\u3059\u7b11<\/p>\n\n\n\n

\u30fbPyTorch + CUDA 11.7 \u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb
\u516c\u5f0f\u306b\u8a18\u8f09\u304c\u3042\u3063\u305f\u3082\u306e\u3092\u305d\u306e\u307e\u307e\u5b9f\u884c\u3057\u307e\u3057\u305f\u3002<\/p>\n\n\n\n

$ conda install pytorch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia<\/p>\n\n\n\n

\u6b63\u76f4\u3053\u306e\u30a4\u30f3\u30b9\u30bf\u30f3\u30b9\u3067\u306f\u3053\u308c\u4e00\u629e\u306a\u306e\u3067\u306f…\u3044\u3084\u3001\u7d50\u679c\u79c1\u306b\u306f\u3053\u308c\u4e00\u629e\u3068\u3044\u3046\u3053\u3068\u3067\u3002<\/p>\n\n\n\n

\u30fbMKL\u306e\u4f9d\u5b58\u56de\u907f
\u3053\u3061\u3089\u3082\u7406\u7531\u3092\u805e\u304b\u308c\u3066\u3082\u56f0\u308b\u306e\u3067\u3059\u304c\u3001RVC\u3060\u304b\u306e\u74b0\u5883\u69cb\u7bc9\u3092\u8a66\u3057\u3066\u3044\u305f\u6642\u306b\u5fc5\u8981\u3067\u3001\u305d\u306e\u4ed6\u306e\u30a2\u30d7\u30ea\u5c0e\u5165\u6642\u3082\u5fc5\u8981\u3060\u3063\u305f\u306e\u3067\u3001\u79c1\u306e\u74b0\u5883\u3067\u306f\u5fc5\u9808\u306a\u306e\u304b\u3082\u3002<\/p>\n\n\n\n

$ conda install \"mkl<2024.1\" \"intel-openmp<2024.1\"<\/code><\/pre>\n\n\n\n
\u30fb\u52d5\u4f5c\u78ba\u8a8d<\/p>\n\n\n\n
python - << 'EOF'\nimport torch\n\nprint(\"Torch version :\", torch.__version__)\nprint(\"CUDA runtime  :\", torch.version.cuda)\nprint(\"CUDA available:\", torch.cuda.is_available())\nif torch.cuda.is_available():\n    print(\"GPU:\", torch.cuda.get_device_name(0))\nEOF<\/code><\/pre>\n\n\n\n
\u3067\u3001
CUDA available: True
\u3068\u304b
GPU: NVIDIA RTX A4000
\u304c\u8868\u793a\u3055\u308c\u3066\u5165\u308c\u3070\u5927\u4e08\u592b\u3001\u306a\u306f\u305a\u3002

\u5171\u901a\u306e\u74b0\u5883\u69cb\u7bc9\u306f\u3053\u3053\u307e\u3067\u3002\u4eee\u60f3\u74b0\u5883\u306e\u5165\u308c\u7269\u306e\u540d\u524d(\u3053\u3053\u3067\u306fdenoiserenv)\u3092\u5909\u3048\u3066\u3084\u308c\u3070\u3053\u306e\u6d41\u308c\u3067\u74b0\u5883\u69cb\u7bc9OK\u3067\u3059\u3002

\u201dOK\u3067\u3059\u201d\u3068\u304b\u66f8\u304f\u3068\u30d7\u30ed\u306e\u7686\u3055\u3093\u306b\u306f\u9f3b\u3067\u7b11\u308f\u308c\u305d\u3046\u3067\u3059\u304c\u3002\u3002\u3002\u6700\u7d42\u7684\u306b\u306f
\u300cMiniconda + PyTorch 2.0.1 + CUDA 11.7 \u3092\u516c\u5f0f\u624b\u9806\u3067\u5165\u308c\u308b\u300d<\/strong>
\u3053\u308c\u3060\u3051\u3067 GPU \u304c\u5b89\u5b9a\u3057\u3066\u4f7f\u3048\u308b\u3088\u3046\u306b\u306a\u308a\u307e\u3057\u305f\u3002

Denoiser\u306e\u8a2d\u5b9a\u3084\u8a66\u3057\u305f\u3053\u3068\u306a\u3069\u306f\u6b21\u306e\u6a5f\u4f1a\u306b\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u3053\u306e\u8a18\u4e8b\u3067\u306f\u3001GPUSOROBAN \u3067 GPU \u304c\u4f7f\u3048\u308b PyTorch \u74b0\u5883\u3092\u6700\u77ed\u3067\u69cb\u7bc9\u3059\u308b\u307e\u3067\u306e\u5b9f\u8df5\u30ed\u30b0\u3092\u66f8\u3044\u3066\u3044\u307e\u3059\u3002 \u77e5\u308a\u5408\u3044\u304cYoutube\u306b\u6295\u7a3f\u3057\u3066\u3044\u305f\u308a\u3001\u81ea\u5206\u3067\u3082\u3084\u3063\u3066\u307f\u305f\u3044\u304b\u306a\u3001\u3068\u601d\u3046\u3053\u3068\u3082\u3042\u3063\u3066\u3001\u4f8b\u3048\u3070 […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3],"tags":[],"class_list":["post-6","post","type-post","status-publish","format-standard","hentry","category-tech"],"_links":{"self":[{"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/posts\/6","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.em99.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6"}],"version-history":[{"count":6,"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/posts\/6\/revisions"}],"predecessor-version":[{"id":47,"href":"https:\/\/www.em99.tech\/index.php?rest_route=\/wp\/v2\/posts\/6\/revisions\/47"}],"wp:attachment":[{"href":"https:\/\/www.em99.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.em99.tech\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.em99.tech\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}