\nNvidia H100 GPU \u3092\u642d\u8f09\u3057\u305f RHEL8 \u3067 MIG \u3092\u6709\u52b9\u5316\u3057\u305f\u306e\u3067\u30e1\u30e2\u3002 <\/p>\n
\u307e\u305a\u306f Nvidia \u30c9\u30e9\u30a4\u30d0\u30fc\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb<\/p>\n
\r\nsudo dnf update -y\r\nsudo dnf install kernel-devel kernel-headers gcc make -y\r\nsudo dnf config-manager --add-repo https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/rhel8\/x86_64\/cuda-rhel8.repo\r\nsudo dnf -y install nvidia-driver-latest-dkms\r\n<\/pre>\n.bash_profile \u306b\u4ee5\u4e0b\u3092\u8ffd\u8a18\u3057\u3066\u304a\u304f\u3002<\/p>\n
\r\nexport PATH=\/usr\/local\/bin:$PATH\r\nexport PATH=\/usr\/local\/cuda\/bin:$PATH\r\nexport LD_LIBRARY_PATH=\/usr\/local\/cuda\/lib64:$LD_LIBRARY_PATH\r\nexport TF_FORCE_GPU_ALLOW_GROWTH=true\r\n<\/pre>\n\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb\u3057\u305f\u3089\u518d\u8d77\u52d5<\/p>\n
\r\nsudo reboot\r\n<\/pre>\n\u3064\u3065\u3044\u3066 CUDA \u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb<\/p>\n
\r\nsudo dnf install -y https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/rhel8\/x86_64\/cuda-keyring-1.0-1.el8.noarch.rpm\r\nsudo dnf config-manager --add-repo https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/rhel8\/x86_64\/cuda-rhel8.repo\r\nsudo dnf install -y cuda-drivers\r\n<\/pre>\n\u305d\u3057\u3066 MIG \u3092\u6709\u52b9\u5316<\/p>\n
\r\n# Enable persistence mode\r\nsudo nvidia-smi -pm 1\r\n\r\n# Enable MIG mode\r\nsudo nvidia-persistenced --persistence-mode\r\nsudo nvidia-smi -mig 1\r\n<\/pre>\n\u305d\u3057\u305f\u3089 MIG \u306e GPU instance profile \u3092\u78ba\u8a8d<\/p>\n
\r\n$ sudo nvidia-smi mig -lgip\r\n+-----------------------------------------------------------------------------+\r\n| GPU instance profiles: |\r\n| GPU Name ID Instances Memory P2P SM DEC ENC |\r\n| Free\/Total GiB CE JPEG OFA |\r\n|=============================================================================|\r\n| 0 MIG 1g.12gb 19 7\/7 10.75 No 16 1 0 |\r\n| 1 1 0 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 1g.12gb+me 20 1\/1 10.75 No 16 1 0 |\r\n| 1 1 1 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 1g.24gb 15 4\/4 21.62 No 26 1 0 |\r\n| 1 1 0 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 2g.24gb 14 3\/3 21.62 No 32 2 0 |\r\n| 2 2 0 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 3g.47gb 9 2\/2 46.38 No 60 3 0 |\r\n| 3 3 0 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 4g.47gb 5 1\/1 46.38 No 64 4 0 |\r\n| 4 4 0 |\r\n+-----------------------------------------------------------------------------+\r\n| 0 MIG 7g.94gb 0 1\/1 93.12 No 132 7 0 |\r\n| 8 7 1 |\r\n+-----------------------------------------------------------------------------+\r\n<\/pre>\n\u3068\u308a\u3042\u3048\u305a 1g.12gb \u306e Compute instance \u3092 7\u500b\u4f5c\u308b\u3002<\/p>\n
\r\n$ sudo nvidia-smi mig -cgi 1g.12gb,1g.12gb,1g.12gb,1g.12gb,1g.12gb,1g.12gb,1g.12gb -C\r\nSuccessfully created GPU instance ID 13 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 13 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 11 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 11 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 12 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 12 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 7 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 7 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 8 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 8 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 9 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 9 using profile MIG 1g.12gb (ID 0)\r\nSuccessfully created GPU instance ID 10 on GPU 0 using profile MIG 1g.12gb (ID 19)\r\nSuccessfully created compute instance ID 0 on GPU 0 GPU instance ID 10 using profile MIG 1g.12gb (ID 0)\r\n<\/pre>\n\u51fa\u6765\u4e0a\u304c\u3063\u305f\u304b\u78ba\u8a8d\u3059\u308b\u3002 7 \u500b\u51fa\u6765\u3066\u305f\u3002<\/p>\n
\r\n$ sudo nvidia-smi mig -lgi\r\n+-------------------------------------------------------+\r\n| GPU instances: |\r\n| GPU Name Profile Instance Placement |\r\n| ID ID Start:Size |\r\n|=======================================================|\r\n| 0 MIG 1g.12gb 19 7 0:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 8 1:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 9 2:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 10 3:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 11 4:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 12 5:1 |\r\n+-------------------------------------------------------+\r\n| 0 MIG 1g.12gb 19 13 6:1 |\r\n+-------------------------------------------------------+\r\n<\/pre>\n\u3053\u308c\u3067\u6e96\u5099\u5b8c\u4e86\u3002
\nPython \u3067\u52d5\u4f5c\u78ba\u8a8d\u3059\u308b\u3002<\/p>\nvenv\u4f5c\u308a\u307e\u3059\u3002<\/p>\n
\r\npython3.11 -m venv ~\/python311_tf_env\r\n\r\n# Activate the virtual environment\r\nsource ~\/python311_tf_env\/bin\/activate\r\n\r\n# Upgrade pip first and install packages\r\npip install --upgrade pip\r\npip install tensorflow[and-cuda]\r\npip install matplotlib\r\npip install scikit-learn\r\n<\/pre>\n\u307e\u305a\u306f MIG \u3092\u8a8d\u8b58\u3067\u304d\u3066\u308b\u304b\u306e\u78ba\u8a8d\u306e\u305f\u3081\u306e python<\/p>\n
\r\nimport os\r\nimport tensorflow as tf\r\n\r\n# Enable memory growth to avoid allocating all GPU memory at once\r\nphysical_devices = tf.config.list_physical_devices('GPU')\r\nprint("Physical devices:", physical_devices)\r\n\r\nif physical_devices:\r\n try:\r\n for gpu in physical_devices:\r\n tf.config.experimental.set_memory_growth(gpu, True)\r\n print("Memory growth enabled")\r\n except Exception as e:\r\n print(f"Error setting memory growth: {e}")\r\n\r\n# Print detailed GPU info\r\nprint("TensorFlow version:", tf.__version__)\r\nprint("CUDA visible devices:", os.environ.get('CUDA_VISIBLE_DEVICES', 'Not set'))\r\n\r\n# Try a simple GPU operation\r\ntry:\r\n with tf.device('\/GPU:0'):\r\n a = tf.constant([[1.0, 2.0], [3.0, 4.0]])\r\n b = tf.constant([[5.0, 6.0], [7.0, 8.0]])\r\n c = tf.matmul(a, b)\r\n print("Matrix multiplication result:", c.numpy())\r\n print("GPU operation successful!")\r\nexcept Exception as e:\r\n print(f"GPU operation failed: {e}")\r\n<\/pre>\n\u5b9f\u884c\u7d50\u679c<\/p>\n
\r\n$ python checkmig.py\r\nPhysical devices: [PhysicalDevice(name='\/physical_device:GPU:0', device_type='GPU')]\r\nMemory growth enabled\r\nTensorFlow version: 2.19.0\r\nCUDA visible devices: MIG-a9bb5f16-9a29-5172-9fe5-7af5908664bf\r\nMIG-5bb83066-7217-5c42-b99e-fe75efe6d7e5\r\nMIG-1f6a0e31-cc52-5909-b14e-3a742942bd35\r\nMIG-56f1b757-9dfd-58cc-b886-8ebe2c442094\r\nMIG-0da82664-471e-5a6b-9738-4a6acf986175\r\nMIG-2e7ddef9-c2f3-5e0b-965f-f4024769c795\r\nMIG-d3ec2ae4-8d3d-5317-920b-3b8219c3176d\r\nMatrix multiplication result: [[19. 22.]\r\n [43. 50.]]\r\nGPU operation successful!\r\n<\/pre>\n\u3061\u3083\u3093\u3068 MIG instance \u3092\u8a8d\u8b58\u3057\u3066\u308b\u3002
\n\u6b21\u306f\u3001\u4e00\u756a\u76ee\u306e MIG instance \u3092\u4f7f\u3063\u3066 python \u3092\u5b9f\u884c\u3057\u3066\u307f\u308b\u3002<\/p>\n\r\n# TensorFlow and keras\r\nimport tensorflow as tf\r\nimport keras\r\nimport numpy as np\r\nimport matplotlib.pyplot as plt\r\nfrom tensorflow.python.client import device_lib\r\nprint(tf.__version__)\r\n\r\n# GPU available?\r\nprint(device_lib.list_local_devices())\r\n\r\n# load the dataset \r\nfashion_mnist = keras.datasets.fashion_mnist\r\n(train_images, train_labels), (test_images, test_labels) = fashion_mnist.load_data()\r\n\r\nclass_names = ['T-shirt\/top', 'Trouser', 'Pullover', 'Dress', 'Coat', \r\n 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot']\r\n\r\ntrain_images = train_images \/ 255.0\r\ntest_images = test_images \/ 255.0\r\n\r\n# define and train the model\r\nmodel = keras.Sequential([\r\n keras.layers.Flatten(input_shape=(28, 28)),\r\n keras.layers.Dense(8192, activation=tf.nn.relu),\r\n keras.layers.Dense(4096, activation=tf.nn.softmax)\r\n])\r\n\r\nmodel.compile(optimizer='adam',\r\n loss='sparse_categorical_crossentropy',\r\n metrics=['accuracy'])\r\n\r\nmodel.fit(train_images, train_labels, epochs=5)\r\n<\/pre>\n\u5b9f\u884c\u7d50\u679c\u3002\u3061\u3083\u3093\u3068 GPU \u4f7f\u3063\u3066\u308b\u304b\u3089\u901f\u3044\u3002<\/p>\n
\r\n$ CUDA_VISIBLE_DEVICES=MIG-a9bb5f16-9a29-5172-9fe5-7af5908664bf python gputest.py\r\n2.19.0\r\n[name: "\/device:CPU:0"\r\ndevice_type: "CPU"\r\nmemory_limit: 268435456\r\nlocality {\r\n}\r\nincarnation: 7443360534658311321\r\nxla_global_id: -1\r\n, name: "\/device:GPU:0"\r\ndevice_type: "GPU"\r\nmemory_limit: 9541976064\r\nlocality {\r\n bus_id: 1\r\n links {\r\n }\r\n}\r\nincarnation: 7883565882498998567\r\nphysical_device_desc: "device: 0, name: NVIDIA H100 NVL MIG 1g.12gb, pci bus id: 0001:00:00.0, compute capability: 9.0"\r\nxla_global_id: 416903419\r\n]\r\nEpoch 1\/5\r\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 10s 5ms\/step - accuracy: 0.7798 - loss: 0.6827\r\nEpoch 2\/5\r\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 9s 5ms\/step - accuracy: 0.8595 - loss: 0.3803\r\nEpoch 3\/5\r\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 9s 5ms\/step - accuracy: 0.8797 - loss: 0.3253\r\nEpoch 4\/5\r\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 9s 5ms\/step - accuracy: 0.8930 - loss: 0.2903\r\nEpoch 5\/5\r\n1875\/1875 \u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501\u2501 9s 5ms\/step - accuracy: 0.8960 - loss: 0.2788\r\n<keras.src.callbacks.history.History at 0x7f9cdc376350>\r\n<\/pre>\n\u4ee5\u4e0a
\n <\/p>\n","protected":false},"excerpt":{"rendered":"Nvidia H100 GPU \u3092\u642d\u8f09\u3057\u305f RHEL8 \u3067 MIG \u3092\u6709\u52b9\u5316\u3057\u305f\u306e\u3067\u30e1\u30e2\u3002 \u307e\u305a\u306f Nvidia \u30c9\u30e9\u30a4\u30d0\u30fc\u306e\u30a4\u30f3\u30b9\u30c8\u30fc\u30eb sudo dnf update -y sudo dnf insta […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[21,8,29],"tags":[],"class_list":["post-2231","post","type-post","status-publish","format-standard","hentry","category-data-engineering","category-linux","category-python"],"_links":{"self":[{"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/posts\/2231","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/comments?post=2231"}],"version-history":[{"count":2,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/posts\/2231\/revisions"}],"predecessor-version":[{"id":2269,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/posts\/2231\/revisions\/2269"}],"wp:attachment":[{"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/media?parent=2231"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/categories?post=2231"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.kwonline.org\/memo2\/wp-json\/wp\/v2\/tags?post=2231"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}