{"id":36143,"date":"2023-01-30T09:43:34","date_gmt":"2023-07-20T03:19:28","guid":{"rendered":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/"},"modified":"2024-04-29T12:13:06","modified_gmt":"2024-04-29T04:13:06","slug":"%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82","status":"publish","type":"post","link":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/","title":{"rendered":"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU"},"content":{"rendered":"

\u5728Kubernetes\u4e0a\u4f7f\u7528GPU\u6765\u8fd0\u884cTensorflow\u3002<\/h1>\n

Kubernetes 1.8\u7248\u672c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b9e\u9a8c\u6027\u529f\u80fd\uff0c\u5141\u8bb8\u4ecePod\u4e2d\u4f7f\u7528GPU\u3002
\n\u4f7f\u7528Kubernetes\u7684GPU\u7684\u597d\u5904\u5728\u4e8e\u53ef\u4ee5\u7ed3\u5408\u6df1\u5ea6\u5b66\u4e60\u548c\u5bb9\u5668\u7f16\u6392\u3002
\n\u5728\u6df1\u5ea6\u5b66\u4e60\u7684\u8bad\u7ec3\u9636\u6bb5\uff0c\u4f7f\u7528GPU\u53ef\u4ee5\u52a0\u5feb\u901f\u5ea6\u3002
\nNvidia\u8fd8\u63d0\u4f9b\u4e86Nvidia Docker\uff0c\u4ee5\u4fbf\u4eceDocker\u4e2d\u4f7f\u7528GPU\u548cCUDA\uff0c\u5e76\u4e14\u5728\u5bb9\u5668\u4e2d\u4f7f\u7528GPU\u5df2\u7ecf\u6210\u4e3a\u4e00\u79cd\u6709\u6548\u7684\u65b9\u6cd5\u3002
\n\u5728Kubernetes\u4e2d\uff0c\u4e5f\u6709\u5bf9GPU\u4f7f\u7528\u7684\u9700\u6c42\uff0c\u5e76\u4e14\u76ee\u524d\u53ef\u4ee5\u5728alpha\u7248\u4e2d\u4f7f\u7528\u3002<\/p>\n

\u4f7f\u7528Kubernetes\u6765\u5229\u7528GPU\u7684\u65b9\u5f0f\u3002<\/h2>\n

\u66f4\u8be6\u7ec6\u7684\u4fe1\u606f\u8bf7\u53c2\u8003\u4ee5\u4e0b\u94fe\u63a5\uff0c\u5176\u4e2d\u63cf\u8ff0\u4e86\u5c06\u5b89\u88c5\u5728\u4e3b\u673a\u670d\u52a1\u5668\u4e0a\u7684Nvidia CUDA\u548cCudnn\u5e93\u6302\u8f7d\u5230Pod\u5e76\u8fdb\u884c\u8c03\u7528\u7684\u673a\u5236\u3002
\nhttps:\/\/kubernetes.io\/docs\/tasks\/manage-gpus\/scheduling-gpus\/<\/p>\n

\u53e6\u5916\uff0c\u8fd8\u6709\u4e00\u4e9b\u5728Kubernetes 1.6\u4e0a\u8bd5\u9a8cKubernetes GPU\u7684\u4f8b\u5b50\uff0c\u8fd9\u4e5f\u53ef\u4ee5\u4f5c\u4e3a\u53c2\u8003\u3002<\/p>\n

\u73af\u5883<\/h2>\n

\u4ee5\u4e0b\u662f\u4f7f\u7528\u7684\u57fa\u7840\u8bbe\u65bd\u548c\u8f6f\u4ef6\uff1a
\n– \u57fa\u7840\u8bbe\u65bd\uff1a\u4e00\u53f0\u8c37\u6b4c\u4e91\u5e73\u53f0\uff08GCP\uff09\u7684Cloud Engine\uff08\u865a\u62df\u673a\uff09
\n– \u89c4\u683c\uff1a4\u4e2aCPU\u300115GB\u5185\u5b58\u3001100GB SSD\u30011\u4e2aGPU\uff08Nvidia K80\uff09
\n– \u64cd\u4f5c\u7cfb\u7edf\uff1aUbuntu16.04
\n– Docker CE 17.03
\n– Kubernetes 1.8
\n– Nvidia CUDA 8.0
\n– Nvidia Cudnn 6.0
\n– Pod\u7684\u5bb9\u5668\u955c\u50cf\uff1agcr.io\/tensorflow\/tensorflow:latest-gpu\uff08\u76ee\u524d\u662fPython2.7\u548cTensorflow1.4\uff09<\/p>\n

\"0.png\"<\/div>\n

\u5404\u4e2a\u8f6f\u4ef6\u7248\u672c\u90fd\u7ecf\u8fc7\u4e86\u6211\u4eec\u57282017\u5e7411\u6708\u5c1d\u8bd5\u7684\u591a\u79cd\u7ec4\u5408\uff0c\u6700\u7ec8\u627e\u5230\u4e86\u4e00\u4e2a\u6709\u6548\u7684\u7ec4\u5408\u3002\u7248\u672c\u5339\u914d\u8d77\u6765\u786e\u5b9e\u5f88\u9ebb\u70e6\u3002<\/p>\n

\u6211\u9009\u62e9\u4f7f\u7528GCP\uff0c\u539f\u56e0\u662f\u865a\u62df\u673a\u90e8\u7f72\u901f\u5ea6\u5feb\u4e14\u6309\u5206\u949f\u8ba1\u8d39\u3002\u7531\u4e8e\u9700\u8981\u591a\u6b21\u5236\u4f5c\u548c\u7834\u574f\uff0c\u4ee5\u4fbf\u7a33\u5b9a\u8fd0\u884c\u4e0d\u540c\u7248\u672c\u7684\u8f6f\u4ef6\uff0c\u6240\u4ee5\u6211\u9700\u8981\u80fd\u591f\u5feb\u901f\u91cd\u5efa\uff0c\u800c\u4e14\u4e0d\u4f1a\u6d6a\u8d39\u91d1\u94b1\u3002<\/p>\n

\u5982\u679c\u6211\u6ca1\u6709\u9519\u7684\u8bdd\uff0cAWS EC2\u662f\u6309\u79d2\u8ba1\u8d39\u7684\uff08\u60ca\u8bb6\uff09\u3002<\/p>\n

\u55ef\uff0c\u65e2\u7136\u8fd9\u6b21\u4f7f\u7528\u7684 Kubernetes \u548c Tensorflow \u90fd\u662f\u8c37\u6b4c\u5f00\u53d1\u7684\uff0c\u5e76\u4e14\u4e91\u7aef\u4e5f\u53ef\u4ee5\u9009\u62e9\u8c37\u6b4c\uff0c\u5e94\u8be5\u662f\u5f88\u597d\u7684\u9009\u62e9\u3002<\/p>\n

\u5efa\u7acb<\/h2>\n

\u5728GCP\u7684Cloud Engine\u4e0a\u90e8\u7f72\u4e00\u53f0Ubuntu16.04\u865a\u62df\u673a\u540e\uff0c\u6211\u4eec\u5c06\u9996\u5148\u8fdb\u884c\u66f4\u65b0\u5e76\u5b89\u88c5Docker CE\uff0c\u7136\u540e\u8fdb\u884c\u5404\u79cd\u914d\u7f6e\u3002\u6709\u5173Docker CE\u7684\u5b89\u88c5\u65b9\u6cd5\uff0c\u8bf7\u53c2\u8003\u4ee5\u4e0b\u94fe\u63a5\uff1ahttps:\/\/docs.docker.com\/engine\/installation\/linux\/docker-ce\/ubuntu\/\u3002\u6709\u5173Kubeadm\u7684\u72ec\u7acb\u5b89\u88c5\uff0c\u8bf7\u53c2\u8003\u4ee5\u4e0b\u94fe\u63a5\uff1ahttps:\/\/kubernetes.io\/docs\/setup\/independent\/install-kubeadm\/\u3002<\/p>\n

# update Linux and install docker<\/span>\r\nsudo <\/span>su -\r\ncurl -fsSL<\/span> https:\/\/download.docker.com\/linux\/ubuntu\/gpg | apt-key add -\r\napt-get -y<\/span> update &&<\/span> apt-get -y<\/span> upgrade\r\n\r\ncurl -fsSL<\/span> https:\/\/download.docker.com\/linux\/ubuntu\/gpg | apt-key add -\r\ncat<\/span> <<<\/span>EOF<\/span> >\/etc\/apt\/sources.list.d\/docker.list\r\ndeb https:\/\/download.docker.com\/linux\/<\/span>$(<\/span>lsb_release -si<\/span> | tr<\/span> '[:upper:]'<\/span> '[:lower:]'<\/span>)<\/span> $(<\/span>lsb_release -cs<\/span>)<\/span> stable\r\n<\/span>EOF\r\n<\/span>apt-get update &&<\/span> apt-get install<\/span> -y<\/span> curl apt-transport-https git vim jq docker-ce=<\/span>$(<\/span>apt-cache madison docker-ce | grep <\/span>17.03 | head<\/span> -1<\/span> | awk<\/span> '{print $3}'<\/span>)<\/span>\r\n\r\n# start docker<\/span>\r\nsystemctl start docker &&<\/span> systemctl enable <\/span>docker\r\n\r\n# set network<\/span>\r\ncat<\/span> <<<\/span> EOF<\/span> > \/tmp\/host\r\nlocalhost\r\n10.10.0.2\r\n<\/span>EOF\r\n<\/span>mkdir<\/span> -p<\/span> ~\/.ssh\/\r\nfor <\/span>i in<\/span> $(<\/span>cat<\/span> \/tmp\/host)<\/span>;<\/span> do <\/span>ssh-keyscan -H<\/span> $i<\/span> >><\/span> ~\/.ssh\/known_hosts ;<\/span> done<\/span>\r\n<\/code><\/pre>\n
\u4e0b\u9762\u662fCUDA\u548cCudnn\u7684\u5b89\u88c5\u8fc7\u7a0b\u3002
\n\u5f53\u5728Kubernetes\u4e0a\u4f7f\u7528GPU\u65f6\uff0c\u9700\u8981\u52a0\u8f7d\u4e3b\u673a\u670d\u52a1\u5668\u7684\u5e93\u6587\u4ef6\uff0c\u56e0\u6b64\u4e3b\u673a\u670d\u52a1\u5668\u9700\u8981\u5b89\u88c5CUDA\u548cCudnn\u3002
\nCUDA\u548cCudnn\u90fd\u53ef\u4ee5\u4f7f\u7528\u7248\u672c\u6307\u5b9a\u8fdb\u884c\u5b89\u88c5\u3002
\n\u8be6\u7ec6\u4fe1\u606f\u8bf7\u8bbf\u95eehttps:\/\/developer.nvidia.com\/cuda-toolkit\u3002<\/p>\n
## find nvidia GPU<\/span>\r\nlspci | grep<\/span> -i<\/span> nvidia\r\n\r\n# install requirements for cuda and cndnn<\/span>\r\nwget https:\/\/developer.download.nvidia.com\/compute\/cuda\/repos\/ubuntu1604\/x86_64\/cuda-repo-ubuntu1604_8.0.61-1_amd64.deb\r\ndpkg -i<\/span> cuda-repo-ubuntu1604_8.0.61-1_amd64.deb\r\napt-get -y<\/span> update\r\napt-get -y<\/span> install<\/span> --allow-unauthenticated<\/span> cuda-8-0\r\n\r\necho<\/span> 'export CUDA_HOME=\/usr\/local\/cuda'<\/span> >><\/span> \/etc\/profile\r\necho<\/span> 'export PATH=\/usr\/local\/cuda-8.0\/bin:${PATH}'<\/span> >><\/span> \/etc\/profile\r\necho<\/span> 'export LD_LIBRARY_PATH=\/usr\/local\/cuda-8.0\/lib64:\/usr\/lib\/nvidia-384:\/usr\/lib\/nvidia:${CUDA_HOME}:${LD_LIBRARY_PATH}'<\/span> >><\/span> \/etc\/profile\r\nsource<\/span> \/etc\/profile\r\n<\/code><\/pre>\n
\u60a8\u53ef\u4ee5\u4f7f\u7528wget\u83b7\u53d6CUDA\u7684\u5b89\u88c5\u7a0b\u5e8f\uff0c\u4f46\u9700\u8981\u5c06Cudnn\u4ece\u6d4f\u89c8\u5668\u4e2d\u4e0b\u8f7d\u5e76\u4e0a\u4f20\u5230\u865a\u62df\u673a\u3002
\n\u8bf7\u5728\u4ee5\u4e0b\u547d\u4ee4\u4e4b\u524d\u4e0b\u8f7dCudnn\u5e76\u4e0a\u4f20\u5230\u865a\u62df\u673a\u3002
\n\u4e0b\u8f7dCudnn\u9700\u8981\u5728Nvidia\u8fdb\u884c\u6ce8\u518c\u5e76\u56de\u7b54\u95ee\u5377\u8c03\u67e5\u3002
\nhttps:\/\/developer.nvidia.com\/rdp\/cudnn-download<\/p>\n
# prerequisite: upload cudnn to the server<\/span>\r\ntar <\/span>zxvf cudnn-8.0-linux-x64-v6.0.tgz\r\ncp <\/span>cuda\/include\/cudnn.h \/usr\/local\/cuda\/include\r\ncp <\/span>cuda\/lib64\/libcudnn*<\/span> \/usr\/local\/cuda\/lib64\r\nchmod <\/span>a+r \/usr\/local\/cuda\/include\/cudnn.h \/usr\/local\/cuda\/lib64\/libcudnn*<\/span>\r\n\r\nnvcc --version<\/span>\r\nnvidia-smi\r\n<\/code><\/pre>\n
\u8bf7\u786e\u8ba4GPU\u5df2\u88ab\u6b63\u786e\u8bc6\u522b\u3002<\/p>\n
\"1.PNG\"<\/div>\n
\u63a5\u4e0b\u6765\u5b89\u88c5Kubernetes\u3002
\n\u4f7f\u7528Kubeadm\u8fdb\u884c\u6784\u5efa\uff0c\u9996\u5148\u5b89\u88c5Kubeadm\u3002
\nhttps:\/\/kubernetes.io\/docs\/setup\/independent\/install-kubeadm\/<\/p>\n
# install kube tools<\/span>\r\napt-get update &&<\/span> apt-get install<\/span> -y<\/span> apt-transport-https\r\ncurl -s<\/span> https:\/\/packages.cloud.google.com\/apt\/doc\/apt-key.gpg | apt-key add -\r\ncat<\/span> <<<\/span>EOF<\/span> >\/etc\/apt\/sources.list.d\/kubernetes.list\r\ndeb http:\/\/apt.kubernetes.io\/ kubernetes-xenial main\r\n<\/span>EOF\r\n\r\n<\/span>apt-get -y<\/span> update\r\napt-get install<\/span> -y<\/span> --allow-unauthenticated<\/span> kubelet kubeadm kubectl kubernetes-cni\r\n<\/code><\/pre>\n
\u4e3a\u4e86\u5728Kubernetes\u4e2d\u4f7f\u7528GPU\uff0c\u9700\u8981\u5728Kubelet\u7684\u542f\u52a8\u547d\u4ee4\u4e2d\u6dfb\u52a0–feature-gates=Accelerators=true\u9009\u9879\u3002
\n\u53ef\u4ee5\u901a\u8fc7\u5c06\u6b64\u9009\u9879\u8ffd\u52a0\u5230\/etc\/systemd\/system\/kubelet.service.d\/\u76ee\u5f55\u4e0b\u7684*-kubeadm.conf\u6587\u4ef6\u4e2d\u6765\u6dfb\u52a0\u8be5\u9009\u9879\u3002<\/p>\n
FILE_NAME<\/span>=<\/span>$(<\/span>ls<\/span> \/etc\/systemd\/system\/kubelet.service.d\/*<\/span>-kubeadm<\/span>.conf)<\/span>\r\nsed<\/span> -i<\/span> '\/^ExecStart=\\\/usr\\\/bin\\\/kubelet\/ s\/$\/ --feature-gates=Accelerators=true\/'<\/span> ${<\/span>FILE_NAME<\/span>}<\/span>\r\n\r\nsystemctl daemon-reload\r\nsystemctl restart kubelet\r\n<\/code><\/pre>\n
\u4f7f\u7528Kubeadm\u642d\u5efaKubernetes\u96c6\u7fa4\u3002
\nhttps:\/\/kubernetes.io\/docs\/setup\/independent\/create-cluster-kubeadm\/<\/p>\n
# initialize kubeadm and kube cluster<\/span>\r\nkubeadm init --pod-network-cidr<\/span>=<\/span>10.244.0.0\/16 --apiserver-advertise-address<\/span>=<\/span>$(<\/span>hostname<\/span> -i<\/span>)<\/span>\r\nsystemctl enable <\/span>kubelet &&<\/span> systemctl start kubelet\r\nsystemctl status kubelet -l<\/span>\r\n\r\nmkdir<\/span> -p<\/span> $HOME<\/span>\/.kube\r\ncp<\/span> -i<\/span> \/etc\/kubernetes\/admin.conf $HOME<\/span>\/.kube\/config\r\nchown<\/span> $(<\/span>id<\/span> -u<\/span>)<\/span>:$(<\/span>id<\/span> -g<\/span>)<\/span> $HOME<\/span>\/.kube\/config\r\n\r\nkubectl cluster-info\r\n\r\n# install flannel<\/span>\r\nkubectl apply -f<\/span> https:\/\/raw.githubusercontent.com\/coreos\/flannel\/v0.9.0\/Documentation\/kube-flannel.yml\r\nkubectl get pods --all-namespaces<\/span>\r\n<\/code><\/pre>\n
\u7531\u4e8e\u672c\u6b21\u4ec5\u914d\u7f6e\u4e00\u4e2a\u4e3b\u670d\u52a1\u5668\uff0c\u6240\u4ee5\u9700\u8981\u542f\u7528\u5bf9\u4e3b\u670d\u52a1\u5668\u7684\u90e8\u7f72\u3002<\/p>\n
# allow deployment to Kubernetes master<\/span>\r\nkubectl taint nodes --all<\/span> node-role.kubernetes.io\/master-\r\n<\/code><\/pre>\n
\u4e3a\u4e86\u786e\u8ba4\u662f\u5426\u8fd0\u4f5c\u6b63\u5e38\uff0c\u6211\u4eec\u5c06\u90e8\u7f72\u4e00\u5bb6\u889c\u5b50\u5e97\u3002<\/p>\n
# deploy sample sock-shop app<\/span>\r\nkubectl create namespace sock-shop\r\nkubectl apply -n<\/span> sock-shop -f<\/span> \"https:\/\/github.com\/microservices-demo\/microservices-demo\/blob\/master\/deploy\/kubernetes\/complete-demo.yaml?raw=true\"<\/span>\r\nkubectl -n<\/span> sock-shop get svc front-end\r\nkubectl get pods -n<\/span> sock-shop\r\n<\/code><\/pre>\n
\u5982\u679c\u80fd\u5920\u9806\u5229\u90e8\u7f72\uff0c\u6211\u6703\u5c07\u5176\u522a\u9664\u3002<\/p>\n
# delete after deployment<\/span>\r\nkubectl delete namespace sock-shop\r\n<\/code><\/pre>\n
\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u5229\u7528GPU\u3002<\/h2>\n
\u8fd9\u6837\u5c31\u5df2\u7ecf\u6784\u5efa\u5b8c\u6210\u4e86\u3002
\n\u7136\u540e\u6211\u4eec\u5c06\u4eceKubernetes\u7684Pod\u4e2d\u4f7f\u7528GPU\u6765\u8fd0\u884cTensorflow\u3002<\/p>\n
Pod\u7684\u5b9a\u4e49yml\u5982\u4e0b\uff1a
\n\u5c06\u73af\u5883\u53d8\u91cf\u8bbe\u7f6e\u4e3aCUDA\u3001Cudnn\u7b49Nvidia GPU\u5e93\u7684\u8def\u5f84\u3002
\n\u53e6\u5916\uff0c\u7531\u4e8eKubernetes\u4f7f\u7528\u4e3b\u673a\u670d\u52a1\u5668\u7684\u5e93\u6765\u8fd0\u884cGPU\uff0c\u6240\u4ee5\u5c06\u4e3b\u673a\u670d\u52a1\u5668\u4e0a\u7684CUDA\u3001Cudnn\u7b49Nvidia GPU\u76f8\u5173\u76ee\u5f55\u6302\u8f7d\u5230\u5bb9\u5668\u4e2d\u3002
\n\uff08\u867d\u7136\u6211\u4eec\u53ef\u80fd\u6302\u8f7d\u4e86\u5f88\u591a\u4e1c\u897f\uff0c\u4f46\u4e5f\u8bb8\u6709\u4e9b\u662f\u4e0d\u5fc5\u8981\u7684\u2026\u2026\uff09<\/p>\n
cat<\/span> <<-<\/span> EOF<\/span> > kubegpu.yml\r\nkind: Pod\r\napiVersion: v1\r\nmetadata:\r\n  name: gpu-pod\r\nspec:\r\n  containers:\r\n  - name: gpu-container\r\n    image: gcr.io\/tensorflow\/tensorflow:latest-gpu\r\n    resources:\r\n      limits:\r\n        alpha.kubernetes.io\/nvidia-gpu: 1\r\n    env:\r\n    - name: CUDA_HOME\r\n      value: '\/usr\/local\/cuda'\r\n    - name: PATH\r\n      value: '\/usr\/local\/cuda-8.0\/bin:<\/span>${<\/span>PATH<\/span>}<\/span>'\r\n    - name: LD_LIBRARY_PATH\r\n      value: '\/usr\/local\/cuda-8.0\/lib64:\/usr\/lib\/nvidia-384:\/usr\/lib\/nvidia:<\/span>${<\/span>CUDA_HOME<\/span>}<\/span>:<\/span>${<\/span>LD_LIBRARY_PATH<\/span>}<\/span>'\r\n    volumeMounts:\r\n    - mountPath: \/usr\/lib\/nvidia-384\/bin\r\n      name: bin\r\n    - mountPath: \/usr\/lib\/nvidia-384\r\n      name: lib\r\n    - mountPath: \/usr\/local\/cuda-8.0\r\n      name: cuda\r\n    - mountPath: \/usr\/lib\/x86_64-linux-gnu\r\n      name: gnu\r\n    - mountPath: \/usr\/bin\/nvidia-smi\r\n      name: nvs\r\n    - mountPath: \/usr\/local\/cuda\/lib64\r\n      name: clib64\r\n    - mountPath: \/usr\/local\/cuda\/include\r\n      name: include\r\n  volumes:\r\n  - hostPath:\r\n      path: \/usr\/lib\/nvidia-384\/bin\r\n    name: bin\r\n  - hostPath:\r\n      path: \/usr\/lib\/nvidia-384\r\n    name: lib\r\n  - hostPath:\r\n      path: \/usr\/local\/cuda-8.0\r\n    name: cuda\r\n  - hostPath:\r\n      path: \/usr\/lib\/x86_64-linux-gnu\r\n    name: gnu\r\n  - hostPath:\r\n      path: \/usr\/bin\/nvidia-smi\r\n    name: nvs\r\n  - hostPath:\r\n      path: \/usr\/local\/cuda\/lib64\r\n    name: clib64\r\n  - hostPath:\r\n      path: \/usr\/local\/cuda\/include\r\n    name: include\r\n<\/span>EOF\r\n<\/span><\/code><\/pre>\n
\u8ba9\u6211\u4eec\u8bd5\u7740\u52a8\u4e00\u4e0b\u5427\u3002<\/p>\n
# create pod<\/span>\r\nkubectl create -f<\/span> kubegpu.yml\r\n\r\n# see if it is running<\/span>\r\nkubectl get pods\r\n<\/code><\/pre>\n
\"2.PNG\"<\/div>\n
\u5f53pod\u542f\u52a8\uff08Running\uff09\u540e\uff0c\u60a8\u53ef\u4ee5\u767b\u5f55\u5230pod\u5e76\u68c0\u67e5\u662f\u5426\u53ef\u4ee5\u8bc6\u522bGPU\u3002<\/p>\n
kubectl exec<\/span> -it<\/span> gpu-pod \/bin\/bash\r\n<\/code><\/pre>\n
nvidia-smi\r\n<\/code><\/pre>\n
\u5982\u679cGPU\u88ab\u8bc6\u522b\u5230\uff0c\u5c06\u4f1a\u663e\u793a\u5982\u4e0b\u5185\u5bb9\u3002<\/p>\n
\"3.PNG\"<\/div>\n
\u8ba9\u6211\u4eec\u6700\u540e\u5c1d\u8bd5\u4e00\u4e0b\u5728Tensorflow\u4e2d\u4f7f\u7528GPU\u3002
\n\u6211\u4eec\u5c06\u4f7f\u7528\u4ee5\u4e0b\u7684\u7a0b\u5e8f\uff1a
\nhttps:\/\/www.tensorflow.org\/tutorials\/using_gpu<\/p>\n
cat<\/span> <<<\/span> EOF<\/span> > ten.py\r\nimport tensorflow as tf\r\n# Creates a graph.\r\na = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')\r\nb = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')\r\nc = tf.matmul(a, b)\r\n# Creates a session with log_device_placement set to True.\r\nsess = tf.Session(config=tf.ConfigProto(log_device_placement=True))\r\n# Runs the op.\r\nprint(sess.run(c))\r\n<\/span>EOF\r\n\r\n<\/span>python ten.py\r\n<\/code><\/pre>\n
\u6211\u6b63\u5728\u8fd0\u884c\u7c7b\u4f3c\u4e0a\u9762\u7b80\u5355\u8ba1\u7b97\u7684\u6d4b\u8bd5\uff0c\u770b\u8d77\u6765GPU\u88ab\u6210\u529f\u8bc6\u522b\u5e76\u4f7f\u7528\u3002<\/p>\n
\"4.PNG\"<\/div>\n
\u6700\u540e<\/h2>\n
\u6211\u80fd\u591f\u4f7f\u7528Kubernetes\u6765\u5229\u7528GPU\u8fd0\u884cTensorFlow\u3002
\n\u6df1\u5ea6\u5b66\u4e60\u8bad\u7ec3\u5f80\u5f80\u662f\u4e34\u65f6\u6027\u7684\u8fd0\u884c\uff0c\u6bcf\u6b21\u4f7f\u7528\u5bb9\u5668\u51c6\u5907\u73af\u5883\u975e\u5e38\u5408\u9002\u3002
\n\u4e00\u65e6\u5f00\u59cb\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\uff0c\u4e3a\u4e86\u521b\u5efa\u4e0d\u540c\u7684\u6a21\u578b\uff0c\u60a8\u5c06\u9700\u8981\u53ca\u65f6\u90e8\u7f72\u5927\u91cf\u7684\u5bb9\u5668\u6765\u6267\u884c\u8bad\u7ec3\u3002
\n\u4e3a\u6b64\uff0c\u50cfKubernetes\u8fd9\u6837\u7684\u5bb9\u5668\u7f16\u6392\u5de5\u5177\u975e\u5e38\u65b9\u4fbf\u3002<\/p>\n
\u4f7f\u7528Nvidia-docker\u6765\u5728\u5bb9\u5668\u4e2d\u8fdb\u884c\u6df1\u5ea6\u5b66\u4e60\u662f\u4e00\u4e2a\u6709\u6548\u7684\u65b9\u6cd5\uff0c\u4f46\u662f\u5f53\u9700\u8981\u7ba1\u7406\u5927\u91cf\u7684\u5bb9\u5668\u65f6\uff0c\u5c31\u9700\u8981\u4f7f\u7528Kubernetes\u6765\u5229\u7528GPU\u3002<\/p>\n
\u672c\u6b21\u4ecb\u7ecd\u4e86\u5b9e\u73b0\u8fd9\u4e24\u79cd\u9700\u6c42\u7684\u65b9\u6cd5\u3002
\n\u5c3d\u7ba1\u5728Kubernetes\u4e2d\u4f7f\u7528GPU\u4ecd\u5904\u4e8e\u5b9e\u9a8c\u6027\u7684Alpha\u7248\u529f\u80fd\u9636\u6bb5\uff0c\u4f46\u5e0c\u671b\u5c3d\u5feb\u8fbe\u5230GA\uff0c\u5e76\u53d8\u5f97\u66f4\u52a0\u65b9\u4fbf\u6613\u7528\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
\u5728Kubernetes\u4e0a\u4f7f\u7528GPU\u6765\u8fd0\u884cTensorflow\u3002 Kubernetes 1.8\u7248\u672c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b9e\u9a8c […]<\/p>\n","protected":false},"author":7,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-36143","post","type-post","status-publish","format-standard","hentry","category-uncategorized"],"yoast_head":"\n\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU - Blog - Silicon Cloud<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.silicloud.com\/zh\/blog\/\u4f7f\u7528kubernetes\u6765\u8fd0\u884ctensorflow\u5e76\u4f7f\u7528gpu\u3002\/\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU\" \/>\n<meta property=\"og:description\" content=\"\u5728Kubernetes\u4e0a\u4f7f\u7528GPU\u6765\u8fd0\u884cTensorflow\u3002 Kubernetes 1.8\u7248\u672c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b9e\u9a8c […]\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.silicloud.com\/zh\/blog\/\u4f7f\u7528kubernetes\u6765\u8fd0\u884ctensorflow\u5e76\u4f7f\u7528gpu\u3002\/\" \/>\n<meta property=\"og:site_name\" content=\"Blog - Silicon Cloud\" \/>\n<meta property=\"article:published_time\" content=\"2023-07-20T03:19:28+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2024-04-29T04:13:06+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/cdn.silicloud.com\/blog-img\/blog\/img\/657d270137434c4406c3a3e5\/7-0.png\" \/>\n<meta name=\"author\" content=\"\u79d1, \u9896\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"\u79d1, \u9896\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"5 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/\",\"name\":\"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU - Blog - Silicon Cloud\",\"isPartOf\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\"},\"datePublished\":\"2023-07-20T03:19:28+00:00\",\"dateModified\":\"2024-04-29T04:13:06+00:00\",\"author\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/8ca01ba7f7362ad4edb7da206a12f29e\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.silicloud.com\/zh\/blog\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#website\",\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/\",\"name\":\"Blog - Silicon Cloud\",\"description\":\"\",\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/8ca01ba7f7362ad4edb7da206a12f29e\",\"name\":\"\u79d1, \u9896\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/8a6fb3cc7ba2f69d2189ba532aec4633ea7ed75ac0af162ec367cb3abc0fb2af?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/8a6fb3cc7ba2f69d2189ba532aec4633ea7ed75ac0af162ec367cb3abc0fb2af?s=96&d=mm&r=g\",\"caption\":\"\u79d1, \u9896\"},\"url\":\"https:\/\/www.silicloud.com\/zh\/blog\/author\/keying\/\"},{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#local-main-organization-logo\",\"url\":\"\",\"contentUrl\":\"\",\"caption\":\"Blog - Silicon Cloud\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU - Blog - Silicon Cloud","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.silicloud.com\/zh\/blog\/\u4f7f\u7528kubernetes\u6765\u8fd0\u884ctensorflow\u5e76\u4f7f\u7528gpu\u3002\/","og_locale":"zh_CN","og_type":"article","og_title":"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU","og_description":"\u5728Kubernetes\u4e0a\u4f7f\u7528GPU\u6765\u8fd0\u884cTensorflow\u3002 Kubernetes 1.8\u7248\u672c\u63d0\u4f9b\u4e86\u4e00\u4e2a\u5b9e\u9a8c […]","og_url":"https:\/\/www.silicloud.com\/zh\/blog\/\u4f7f\u7528kubernetes\u6765\u8fd0\u884ctensorflow\u5e76\u4f7f\u7528gpu\u3002\/","og_site_name":"Blog - Silicon Cloud","article_published_time":"2023-07-20T03:19:28+00:00","article_modified_time":"2024-04-29T04:13:06+00:00","og_image":[{"url":"https:\/\/cdn.silicloud.com\/blog-img\/blog\/img\/657d270137434c4406c3a3e5\/7-0.png"}],"author":"\u79d1, \u9896","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"\u79d1, \u9896","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"5 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/","url":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/","name":"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU - Blog - Silicon Cloud","isPartOf":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website"},"datePublished":"2023-07-20T03:19:28+00:00","dateModified":"2024-04-29T04:13:06+00:00","author":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/8ca01ba7f7362ad4edb7da206a12f29e"},"breadcrumb":{"@id":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.silicloud.com\/zh\/blog\/"},{"@type":"ListItem","position":2,"name":"\u4f7f\u7528Kubernetes\u6765\u8fd0\u884cTensorflow\u5e76\u4f7f\u7528GPU"}]},{"@type":"WebSite","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#website","url":"https:\/\/www.silicloud.com\/zh\/blog\/","name":"Blog - Silicon Cloud","description":"","inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/8ca01ba7f7362ad4edb7da206a12f29e","name":"\u79d1, \u9896","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/8a6fb3cc7ba2f69d2189ba532aec4633ea7ed75ac0af162ec367cb3abc0fb2af?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/8a6fb3cc7ba2f69d2189ba532aec4633ea7ed75ac0af162ec367cb3abc0fb2af?s=96&d=mm&r=g","caption":"\u79d1, \u9896"},"url":"https:\/\/www.silicloud.com\/zh\/blog\/author\/keying\/"},{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.silicloud.com\/zh\/blog\/%e4%bd%bf%e7%94%a8kubernetes%e6%9d%a5%e8%bf%90%e8%a1%8ctensorflow%e5%b9%b6%e4%bd%bf%e7%94%a8gpu%e3%80%82\/#local-main-organization-logo","url":"","contentUrl":"","caption":"Blog - Silicon Cloud"}]}},"_links":{"self":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/36143","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/comments?post=36143"}],"version-history":[{"count":2,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/36143\/revisions"}],"predecessor-version":[{"id":85207,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/posts\/36143\/revisions\/85207"}],"wp:attachment":[{"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/media?parent=36143"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/categories?post=36143"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.silicloud.com\/zh\/blog\/wp-json\/wp\/v2\/tags?post=36143"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}