2.4 \u5bb9\u5668\u5316\u73af\u5883\u6311\u6218<\/h3>\n

\n
• Docker\/Kubernetes\u9a71\u52a8\u900f\u4f20\uff1a\u5bb9\u5668\u8fd0\u884c\u65f6\u65e0\u6cd5\u8bc6\u522bGPU\u8bbe\u5907\u7684\u60c5\u51b5\uff0c\u591a\u56e0\u7f3a\u5c11`nvidia-container-toolkit`\u6216cgroup\u914d\u7f6e\u4e0d\u5f53\u3002<\/li>\n
• \u865a\u62df\u5316\u51b2\u7a81\uff1aKVM\/QEMU\u4e2d\u7684GPU\u900f\u4f20\u9700\u56fa\u4ef6\u652f\u6301\u4e0e\u7cbe\u51c6\u7684PCI\u8bbe\u5907\u5206\u914d\uff0c\u8f7b\u5fae\u7684\u9a71\u52a8\u7248\u672c\u53d8\u66f4\u5c31\u53ef\u80fd\u5bfc\u81f4\u529f\u80fd\u5931\u6548\u3002<\/li>\n<\/ul>\n

  [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n

  3. \u517c\u5bb9\u6027\u95ee\u9898\u7684\u56db\u6b65\u8bca\u65ad\u6d41\u7a0b<\/strong><\/h2>\n

  \u6709\u6761\u7406\u7684\u68c0\u6d4b\u662f\u51c6\u786e\u5b9a\u4f4d\u95ee\u9898\u7684\u5173\u952e\uff0c\u9075\u5faa\u4ee5\u4e0b\u7ed3\u6784\u5316\u6d41\u7a0b\uff1a<\/p>\n

  3.1 \u83b7\u53d6\u786c\u4ef6\u4fe1\u606f<\/h3>\n

  \n
  1. \u901a\u8fc7\u7ec8\u7aef\u547d\u4ee4\u8bc6\u522bGPU\u578b\u53f7\uff1a\n

     lspci | grep -i vga\r\n            # NVIDIA\u4e13\u5c5e\u4fe1\u606f\u67e5\u8be2\uff1anvidia-smi -L<\/code><\/pre>\n<\/li>\n
     \u901a\u8fc7\u670d\u52a1\u5668\u7ba1\u7406\u9762\u677f\uff08\u5982Dell iDRAC\u3001HPE iLO\uff09\u4ea4\u53c9\u9a8c\u8bc1\uff0c\u786e\u8ba4\u7269\u7406GPU\u5b58\u5728\u72b6\u6001\u4e0e\u56fa\u4ef6\u7248\u672c\u3002<\/li>\n<\/ol>\n
     3.2 \u68c0\u67e5\u7cfb\u7edf\u73af\u5883\u8be6\u60c5<\/h3>\n
     \n
     \u5185\u6838\u7248\u672c\uff1a`uname -r`\uff08\u9a71\u52a8\u6a21\u5757\u517c\u5bb9\u6027\u7684\u5173\u952e\u4f9d\u636e\uff09<\/li>\n
     Xorg\u670d\u52a1\u5668\u7248\u672c\uff1a`Xorg -version`\uff08\u9700\u4e0e\u9a71\u52a8\u6587\u6863\u8981\u6c42\u7684\u7248\u672c\u5339\u914d\uff09<\/li>\n
     Linux\u53d1\u884c\u7248\u4fe1\u606f\uff1a`lsb_release -a`\uff08\u5305\u7ba1\u7406\u5668\u5b89\u88c5\u65b9\u5f0f\u7684\u6838\u5fc3\u53c2\u8003\uff09<\/li>\n<\/ol>\n
     3.3 \u9a8c\u8bc1\u9a71\u52a8\u5b89\u88c5\u72b6\u6001<\/h3>\n
     \n
     NVIDIA\uff1a\u6267\u884c`nvidia-smi`\u2014\u2014\u65e0\u8f93\u51fa\u5373\u8868\u793a\u5b89\u88c5\u5931\u8d25\u6216\u6a21\u5757\u52a0\u8f7d\u5f02\u5e38\u3002<\/li>\n
     AMD\uff1a\u901a\u8fc7`amdgpu-pro –list`\u67e5\u770b\u5df2\u5b89\u88c5\u9a71\u52a8\u7248\u672c\uff1b\u901a\u8fc7`lsmod | grep nouveau`\u53ef\u68c0\u6d4b\u4e0e\u5f00\u6e90nouveau\u9a71\u52a8\u7684\u51b2\u7a81\u3002<\/li>\n<\/ol>\n
     3.4 \u5206\u6790\u7cfb\u7edf\u65e5\u5fd7<\/h3>\n
     \n
     Xorg\u9519\u8bef\u65e5\u5fd7\uff1a\u67e5\u770b`\/var\/log\/Xorg.0.log`\u4e2d\u542b`EE`\uff08\u9519\u8bef\u6807\u8bc6\uff09\u7684\u884c\uff0c\u5b9a\u4f4dGPU\u521d\u59cb\u5316\u76f8\u5173\u95ee\u9898\u3002<\/li>\n
     \u5185\u6838\u6d88\u606f\uff1a`dmesg | grep -iE ‘nvidia|amd|gpu|vga’`\u53ef\u663e\u793a\u5e95\u5c42\u9a71\u52a8\u52a0\u8f7d\u9519\u8bef\uff0c\u5982\u7f3a\u5931\u56fa\u4ef6 blob \u6216PCIe\u679a\u4e3e\u5931\u8d25\u3002<\/li>\n<\/ol>\n
     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
     4. \u5206\u573a\u666f\u89e3\u51b3\u65b9\u6848<\/strong><\/h2>\n
     4.1 \u57fa\u7840\u9a71\u52a8\u5b89\u88c5\u65b9\u5f0f<\/h3>\n
     \u6839\u636e\u670d\u52a1\u5668\u73af\u5883\uff08\u65e0\u754c\u9762\u3001\u5e26GUI\u3001\u5bb9\u5668\u5316\uff09\u9009\u62e9\u5408\u9002\u7684\u5b89\u88c5\u65b9\u6848\uff1a<\/p>\n
     4.1.1 \u5b98\u65b9\u4e13\u6709\u9a71\u52a8<\/h4>\n
     \n
     NVIDIA\uff08\u65e0\u754c\u9762\u670d\u52a1\u5668\uff09\uff1a\n
     chmod +x NVIDIA-Linux-x86_64-535.54.03.run\r\n            .\/NVIDIA-Linux-x86_64-535.54.03.run --no-x-check --no-nouveau-check --silent<\/code><\/pre>\n
     \u6ce8\uff1a\u5982\u9700\u7981\u7528nouveau\u9a71\u52a8\uff0c\u5148\u6267\u884c`sudo modprobe -r nouveau`\u3002<\/em><\/li>\n
     AMD GPU-Pro\uff08\u4f01\u4e1a\u7ea7\u573a\u666f\uff09\uff1a\n
     sudo apt update && sudo apt install amdgpu-pro-core\r\n            sudo amdgpu-pro --install --no-dkms<\/code><\/pre>\n<\/li>\n<\/ul>\n
     4.1.2 \u5f00\u6e90\u9a71\u52a8\u66ff\u4ee3\u65b9\u6848<\/h4>\n
     \n
     Nouveau\uff08\u975e\u6027\u80fd\u654f\u611f\u573a\u666f\uff09\uff1a\n
     \n
     \u901a\u8fc7\u5185\u6838\u53c2\u6570\u542f\u7528\uff1a\u5728`\/etc\/default\/grub`\u4e2d\u6dfb\u52a0`nouveau.modeset=1`<\/li>\n
     \u91cd\u65b0\u751f\u6210GRUB\u914d\u7f6e\uff1a`sudo update-grub`<\/li>\n<\/ol>\n<\/li>\n
     AMDGPU\uff08\u5f00\u6e90\uff09\uff1a\u591a\u6570\u73b0\u4ee3\u5185\u6838\u5df2\u5185\u7f6e\uff0c\u9700\u786e\u4fdd`linux-firmware`\u5305\u5df2\u66f4\u65b0\u4ee5\u83b7\u5f97\u5b8c\u6574\u786c\u4ef6\u652f\u6301\u3002<\/li>\n<\/ul>\n
     4.1.3 \u5305\u7ba1\u7406\u5668\u5b89\u88c5<\/h4>\n
     \n
     Debian\/Ubuntu\u7cfb\u5217\uff1a`sudo apt install nvidia-driver-535`\uff08\u5c06\u7248\u672c\u53f7\u66ff\u6362\u4e3a\u76ee\u6807\u7248\u672c\uff09<\/li>\n
     Red Hat\/CentOS\u7cfb\u5217\uff1a`sudo dnf install xorg-x11-drv-nvidia`\uff08\u9700\u4f9d\u8d56RPM Fusion\u4ed3\u5e93\u83b7\u53d6\u975e\u514d\u8d39\u9a71\u52a8\uff09<\/li>\n<\/ul>\n
     4.2 \u5185\u6838\u66f4\u65b0\u540e\u7684\u9a71\u52a8\u6062\u590d<\/h3>\n
     \n
     \u91cd\u65b0\u751f\u6210initramfs\uff1aArch Linux\u7cfb\u7edf\u6267\u884c`sudo mkinitcpio -P`\uff0cDebian\u7cfb\u5217\u6267\u884c`sudo update-initramfs -u`<\/li>\n
     \u91cd\u65b0\u914d\u7f6eGRUB\uff1a\u591a\u5f15\u5bfc\u73af\u5883\u9700\u6267\u884c\u6b64\u6b65\u9aa4\uff0c\u786e\u4fdd\u65b0\u5185\u6838\u52a0\u8f7d\u6b63\u786e\u7684\u9a71\u52a8\u6a21\u5757\u3002<\/li>\n
     \u90e8\u7f72DKMS\uff1a\u901a\u8fc7`sudo apt install dkms`\u5b89\u88c5\u52a8\u6001\u5185\u6838\u6a21\u5757\u652f\u6301\uff0c\u5b9e\u73b0\u5185\u6838\u66f4\u65b0\u65f6\u81ea\u52a8\u91cd\u5efa\u9a71\u52a8\u6a21\u5757\u3002<\/li>\n<\/ol>\n
     4.3 \u4f9d\u8d56\u51b2\u7a81\u89e3\u51b3<\/h3>\n
     \n
     \u7248\u672c\u9501\u5b9a\uff1a\u5728Debian\u7cfb\u7edf\u4e2d\u4f7f\u7528`apt-mark hold nvidia-driver-535`\uff0c\u9632\u6b62\u81ea\u52a8\u5347\u7ea7\u7834\u574f\u517c\u5bb9\u6027\u3002<\/li>\n
     \u624b\u52a8\u89e3\u51b3\u4f9d\u8d56\uff1a\u4ece\u5382\u5546\u4ed3\u5e93\u4e0b\u8f7d\u7279\u5b9a.deb\u6216.rpm\u5305\uff0c\u901a\u8fc7`dpkg -i`\u5b89\u88c5\u3002<\/li>\n
     \u5f7b\u5e95\u5378\u8f7d\u6b8b\u7559\uff1a\u6267\u884c`sudo apt purge ‘*nvidia*’ && sudo apt autoremove`\u6e05\u9664\u6b8b\u7559\u9a71\u52a8\uff0c\u518d\u8fdb\u884c\u5168\u65b0\u5b89\u88c5\u3002<\/li>\n<\/ul>\n
     4.4 \u5bb9\u5668\u4e0e\u865a\u62df\u5316\u4fee\u590d<\/h3>\n
     \n
     Docker GPU\u652f\u6301\u914d\u7f6e\uff1a\n
     sudo apt install nvidia-container-toolkit\r\n            docker run --gpus all --rm nvidia\/cuda:12.0-base nvidia-smi<\/code><\/pre>\n<\/li>\n
     Kubernetes\u8bbe\u5907\u63d2\u4ef6\uff1a\n
     \n
     \u901a\u8fc7DaemonSet\u90e8\u7f72NVIDIA\u8bbe\u5907\u63d2\u4ef6\uff1a\n
     kubectl apply -f https:\/\/raw.githubusercontent.com\/NVIDIA\/k8s-device-plugin\/v0.14.3\/nvidia-device-plugin.yml<\/code><\/pre>\n<\/li>\n
     \u4e3aGPU\u8282\u70b9\u914d\u7f6e\u6c61\u70b9\u4e0e\u5bb9\u5fcd\u7b56\u7565\uff0c\u786e\u4fddPod\u6b63\u786e\u8c03\u5ea6\u3002<\/li>\n<\/ol>\n<\/li>\n<\/ul>\n
     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
     5. \u7f8e\u56fd\u670d\u52a1\u5668\u4e13\u5c5e\u4f18\u5316\u5efa\u8bae<\/strong><\/h2>\n
     \u7f8e\u56fd\u7684\u6570\u636e\u4e2d\u5fc3\u4e0e\u4e91\u73af\u5883\u901a\u5e38\u5177\u5907\u72ec\u7279\u7684\u57fa\u7840\u8bbe\u65bd\u9700\u6c42\uff0c\u9700\u9488\u5bf9\u6027\u4f18\u5316\uff1a<\/p>\n
     5.1 \u6570\u636e\u4e2d\u5fc3\u89c4\u6a21\u90e8\u7f72<\/h3>\n
     \n
     \u6279\u91cf\u5b89\u88c5\u811a\u672c\uff1a\u4f7f\u7528Ansible Playbook\u6216Chef\u914d\u65b9\u5b9e\u73b0\u6570\u767e\u53f0\u670d\u52a1\u5668\u7684\u9a71\u52a8\u6279\u91cf\u90e8\u7f72\uff1a\n
     - name: \u5b89\u88c5NVIDIA\u9a71\u52a8\r\n              become: yes\r\n              command: .\/NVIDIA-Linux-x86_64-{{ driver_version }}.run --silent<\/code><\/pre>\n<\/li>\n
     \u65e0\u754c\u9762IPMI\u914d\u7f6e\uff1a\u901a\u8fc7\u8fdc\u7a0bKVM\u6302\u8f7d\u9a71\u52a8ISO\uff0c\u65e0\u9700\u672c\u5730\u63a7\u5236\u53f0\u5373\u53ef\u6267\u884c\u5b89\u88c5\u64cd\u4f5c\u3002<\/li>\n<\/ul>\n
     5.2 \u4e91\u670d\u52a1\u5668\u6ce8\u610f\u4e8b\u9879<\/h3>\n
     \n
     AWS\/GCP\/Azure\u5e73\u53f0\u5dee\u5f02\uff1a\n
     \n
     AWS EC2\uff1a\u4f7f\u7528NVIDIA\u4f18\u5316AMI\uff0c\u6216\u901a\u8fc7`nvidia-accelerated-image`\u811a\u672c\u5b89\u88c5\u9a71\u52a8\u3002<\/li>\n
     GCP\u8ba1\u7b97\u5f15\u64ce\uff1a\u5728\u9879\u76ee\u63a7\u5236\u53f0\u542f\u7528GPU API\uff0c\u6216\u4f7f\u7528\u6df1\u5ea6\u5b66\u4e60VM\u4e2d\u7684\u9884\u5b89\u88c5\u9a71\u52a8\u3002<\/li>\n<\/ul>\n<\/li>\n
     \u4e91\u539f\u751f\u5de5\u5177\u96c6\uff1a\u501f\u52a9NVIDIA Cloud Native Toolkit\u5b9e\u73b0Kubernetes\u73af\u5883\u4e0b\u7684GPU\u8d44\u6e90\u7ba1\u7406\u3002<\/li>\n<\/ul>\n
     5.3 \u4e3b\u52a8\u76d1\u63a7\u65b9\u6848<\/h3>\n
     \n
     \u7f16\u5199\u9a71\u52a8\u5065\u5eb7\u68c0\u67e5\u811a\u672c\uff1a\n
     while true; do\r\n                nvidia-smi --query-gpu=driver_version,name,utilization.gpu,memory.used --format=csv,noheader\r\n                sleep 3600\r\n            done | tee gpu_monitor.log<\/code><\/pre>\n<\/li>\n
     \u96c6\u6210\u76d1\u63a7\u5de5\u5177\uff1a\u5f53`nvidia-smi`\u8fd4\u56de\u975e\u96f6\u9000\u51fa\u7801\u65f6\uff0c\u901a\u8fc7Prometheus\/Grafana\u53d1\u9001\u544a\u8b66\u3002<\/li>\n<\/ul>\n
     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
     6. \u9884\u9632\u63aa\u65bd\u4e0e\u6700\u4f73\u5b9e\u8df5<\/strong><\/h2>\n
     6.1 \u786c\u4ef6\u91c7\u8d2d\u524d\u7684\u5c3d\u804c\u8c03\u67e5<\/h3>\n
     \n
     \u67e5\u9605\u5382\u5546\u517c\u5bb9\u6027\u5217\u8868\uff1a\n
     \n
     NVIDIA\uff1aLinux\u9a71\u52a8\u652f\u6301\u77e9\u9635<\/li>\n
     AMD\uff1aGPU Linux\u9a71\u52a8\u652f\u6301\u6587\u6863<\/li>\n<\/ul>\n<\/li>\n
     \u9009\u62e9\u5728\u7f8e\u56fd\u6709\u670d\u52a1\u652f\u6301\u3001\u4e14Linux\u517c\u5bb9\u6027\u8bb0\u5f55\u826f\u597d\u7684\u786c\u4ef6\u5382\u5546\uff0c\u5c24\u5176\u9488\u5bf9NVIDIA A100\u3001AMD MI200\u7b49\u4f01\u4e1a\u7ea7GPU\u3002<\/li>\n<\/ol>\n
     6.2 \u9a71\u52a8\u7248\u672c\u7ba1\u7406<\/h3>\n
     \n
     \u7248\u672c\u9501\u5b9a\uff1a\u901a\u8fc7`dpkg –set-selections`\u9632\u6b62\u610f\u5916\u5347\u7ea7\uff1a\n
     echo \"nvidia-driver-535 hold\" | sudo dpkg --set-selections<\/code><\/pre>\n<\/li>\n
     \u5efa\u7acb\u6d4b\u8bd5\u6d41\u6c34\u7ebf\uff1a\u5728\u9884\u53d1\u73af\u5883\u9a8c\u8bc1\u9a71\u52a8\u66f4\u65b0\u540e\uff0c\u518d\u90e8\u7f72\u5230\u751f\u4ea7\u96c6\u7fa4\u3002<\/li>\n<\/ul>\n
     6.3 \u7cfb\u7edf\u5316\u5185\u6838\u4e0e\u8f6f\u4ef6\u5347\u7ea7<\/h3>\n
     \n
     \u91c7\u7528\u5185\u6838\u5c0f\u7248\u672c\u5347\u7ea7\u7b56\u7565\uff1a\u5148\u901a\u8fc7`linux-image-$(uname -r | sed ‘s\/-[0-9]\\+\/\/’)-generic-lts`\u6d4b\u8bd5\uff0c\u518d\u5168\u9762\u90e8\u7f72\u3002<\/li>\n
     \u7248\u672c\u540c\u6b65\uff1a\u59cb\u7ec8\u901a\u8fc7\u5382\u5546\u63d0\u4f9b\u7684\u5de5\u5177\u94fe\uff0c\u540c\u6b65\u66f4\u65b0CUDA\/CuDNN\u4e0eGPU\u9a71\u52a8\u3002<\/li>\n<\/ul>\n
     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
     7. \u7591\u96be\u95ee\u9898\u8fdb\u9636\u6392\u67e5<\/strong><\/h2>\n
     7.1 \u663e\u793a\u5f02\u5e38\uff08\u9ed1\u5c4f\/\u82b1\u5c4f\uff09<\/h3>\n
     \n
     \u8fdb\u5165\u6551\u63f4\u6a21\u5f0f\uff1a\u901a\u8fc7`systemctl rescue.target`\u542f\u52a8\uff0c\u907f\u514dXorg\u5e72\u6270\u6392\u67e5\u3002<\/li>\n
     \u9a71\u52a8\u7b7e\u540d\u95ee\u9898\uff1a\u5728BIOS\u4e2d\u7981\u7528Secure Boot\uff0c\u6216\u4ece\u786c\u4ef6\u5382\u5546\u83b7\u53d6\u5df2\u7b7e\u540d\u9a71\u52a8\u3002<\/li>\n<\/ol>\n
     7.2 \u6027\u80fd\u4e0b\u964d<\/h3>\n
     \n
     \u6027\u80fd\u5206\u6790\u5de5\u5177\uff1a\u4f7f\u7528NVIDIA Nsight Systems\u6216AMD ROCm Profiler\u5b9a\u4f4d\u9a71\u52a8\u5c42\u74f6\u9888\u3002<\/li>\n
     \u5185\u5b58\u6cc4\u6f0f\u68c0\u6d4b\uff1a\u901a\u8fc7`nvidia-smi –loop 10`\u76d1\u63a7\u7a7a\u95f2\u8fdb\u7a0b\u7684\u5185\u5b58\u5360\u7528\u53d8\u5316\uff0c\u8bc6\u522b\u6f5c\u5728\u9a71\u52a8\u6f0f\u6d1e\u3002<\/li>\n<\/ul>\n
     7.3 \u5229\u7528\u793e\u533a\u8d44\u6e90<\/h3>\n
     \n
     \u5b98\u65b9\u8bba\u575b\uff1a\u5728NVIDIA Developer Forums\u6216AMD Community\u83b7\u53d6\u5382\u5546\u4e13\u5c5e\u6280\u672f\u652f\u6301\u3002<\/li>\n
     Wiki\u8d44\u6e90\uff1a\u53c2\u8003Arch Linux NVIDIA Wiki\u83b7\u53d6\u5e95\u5c42\u914d\u7f6e\u7ec6\u8282\u3002<\/li>\n<\/ul>\n
     [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””]<\/p>\n
     8. \u7ed3\u8bba\uff1a\u6784\u5efa\u9ad8\u53ef\u9760\u7684GPU\u52a0\u901f\u670d\u52a1\u5668\u67b6\u6784<\/strong><\/h2>\n
     Linux\u73af\u5883\u4e0b\u7684GPU\u9a71\u52a8\u517c\u5bb9\u6027\u95ee\u9898\uff08\u5c24\u5176\u5728\u7f8e\u56fd\u670d\u52a1\u5668\u79df\u7528\u4e0e\u670d\u52a1\u5668\u6258\u7ba1\u573a\u666f\u4e2d\uff09\uff0c\u9700\u8981\u7ed3\u5408\u7cfb\u7edf\u5316\u8bca\u65ad\u3001\u5382\u5546\u4e13\u5c5e\u65b9\u6848\u4e0e\u4e3b\u52a8\u7ba1\u7406\u3002\u9075\u5faa\u672c\u6587\u6240\u8ff0\u7684\u7ed3\u6784\u5316\u65b9\u6cd5\u2014\u2014\u4ece\u786c\u4ef6\u68c0\u6d4b\u5230\u8fdb\u9636\u4f18\u5316\uff0c\u6280\u672f\u56e2\u961f\u53ef\u786e\u4fddGPU\u52a0\u901f\u5e94\u7528\u7684\u7a33\u5b9a\u8fd0\u884c\u3002\u968f\u7740\u5bb9\u5668\u5316\u4e0eAI\u8d1f\u8f7d\u5bf9\u670d\u52a1\u5668\u67b6\u6784\u7684\u9700\u6c42\u6301\u7eed\u589e\u957f\uff0c\u638c\u63e1\u8fd9\u4e9b\u517c\u5bb9\u6027\u89e3\u51b3\u65b9\u6848\uff0c\u5c06\u6210\u4e3a\u7ef4\u6301\u9ad8\u6027\u80fd\u4e0e\u9ad8\u53ef\u9760\u6027\u7684\u5173\u952e\u3002<\/p>\n
     \u5efa\u8bae\u7acb\u5373\u5ba1\u8ba1\u670d\u52a1\u5668\u7684GPU\u9a71\u52a8\u72b6\u6001\uff0c\u5e76\u5c06\u672c\u6587\u6536\u85cf\u4e3a\u8fd0\u7ef4\u624b\u518c\u3002\u9047\u5230\u7279\u6b8a\u95ee\u9898\uff1f\u6b22\u8fce\u5728\u8bc4\u8bba\u533a\u5206\u4eab\u7ecf\u9a8c\uff0c\u52a9\u529b\u793e\u533a\u5171\u540c\u6210\u957f\u3002<\/p>\n
     [\/vc_column_text][\/vc_column][\/vc_row]<\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"
     [vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””] 1. \u5f15\u8a00\uff1aGPU\u9a71\u52a8\u517c\u5bb9\u6027\u5728\u670d\u52a1\u5668\u73af\u5883\u4e2d\u7684\u5173\u952e\u4f5c\u7528 \u5728Linux\u670d\u52a1\u5668\u7684\u9ad8\u6027\u80fd\u8ba1\u7b97\u9886\u57df\u2014\u2014\u5c24\u5176\u662f\u6df1\u5ea6\u5b66\u4e60\u3001\u79d1\u5b66\u6a21\u62df\u3001\u56fe\u5f62\u6e32\u67d3\u7b49\u573a\u666f\u4e2d\uff0cGPU\u9a71\u52a8\u7a0b\u5e8f\u517c\u5bb9\u6027\u95ee\u9898\u5e38\u6210\u4e3a\u68d8\u624b\u74f6\u9888\u3002\u5bf9\u4e8e\u8fd0\u8425\u7f8e\u56fd\u672c\u571f\u670d\u52a1\u5668\u79df\u7528\u4e0e\u670d\u52a1\u5668\u6258\u7ba1\u670d\u52a1\u7684\u4eba\u5458\u800c\u8a00\uff0c\u4e0d\u7a33\u5b9a\u6216\u4e0d\u517c\u5bb9\u7684GPU\u9a71\u52a8\u4e0d\u4ec5\u4f1a\u4e2d\u65ad\u5173\u952e\u4e1a\u52a1\u5e94\u7528\uff0c\u8fd8\u4f1a\u524a\u5f31\u670d\u52a1\u5668\u67b6\u6784\u7684\u53ef\u9760\u6027\u3002\u672c\u6587\u6df1\u5165\u63a2\u8ba8\u7cfb\u7edf\u5316\u7684\u8bca\u65ad\u3001\u89e3\u51b3\u4e0e\u9884\u9632\u65b9\u6848\uff0c\u4e13\u4e3a\u7ba1\u7406\u642d\u8f7dNVIDIA\u3001AMD\u3001Intel\u72ec\u7acbGPU\u7684Linux\u73af\u5883\u7684\u6280\u672f\u4eba\u5458\u91cf\u8eab\u6253\u9020\u3002 [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””] 2. \u5e38\u89c1GPU\u9a71\u52a8\u517c\u5bb9\u6027\u95ee\u9898\u7c7b\u578b \u7406\u89e3\u517c\u5bb9\u6027\u95ee\u9898\u7684\u672c\u8d28\u662f\u9ad8\u6548\u89e3\u51b3\u95ee\u9898\u7684\u7b2c\u4e00\u6b65\uff0c\u4ee5\u4e0b\u662f\u6700\u5178\u578b\u7684\u95ee\u9898\u7c7b\u522b\uff1a 2.1 \u9a71\u52a8\u7248\u672c\u4e0e\u5185\u6838\u4e0d\u5339\u914d \u5185\u6838\u66f4\u65b0\u540e\u5931\u6548\uff1a\u5e38\u89c1\u573a\u666f\u4e3aLinux\u5185\u6838\u66f4\u65b0\uff08\u5982\u4ece5.15\u5347\u7ea7\u81f36.0\uff09\u540e\uff0c\u539f\u6b63\u5e38\u8fd0\u884c\u7684NVIDIA\u6216AMD\u9a71\u52a8\u56e0\u5185\u6838\u6a21\u5757API\u53d8\u5316\u800c\u65e0\u6cd5\u4f7f\u7528\u3002 \u67b6\u6784\u51b2\u7a81\uff1a32\u4f4d\u4e0e64\u4f4d\u9a71\u52a8\u4e0d\u5339\u914d\uff0c\u5728\u4ecd\u8fd0\u884c32\u4f4d\u7528\u6237\u7a7a\u95f4\u4e0e64\u4f4d\u5185\u6838\u5e76\u5b58\u7684 legacy \u670d\u52a1\u5668\u73af\u5883\u4e2d\u5c24\u4e3a\u7a81\u51fa\u3002 2.2 \u786c\u4ef6\u5382\u5546\u4e13\u5c5e\u652f\u6301\u7f3a\u53e3 NVIDIA\uff1a\u867d\u4e3a\u73b0\u4ee3GPU\u63d0\u4f9b\u5b8c\u5584\u7684Linux\u652f\u6301\uff0c\u4f46GeForce 600\u7cfb\u5217\u7b49\u65e7\u578b\u53f7\u5728\u7279\u5b9a\u5185\u6838\u7248\u672c\u540e\u53ef\u80fd\u4e0d\u518d\u83b7\u5f97\u5b98\u65b9\u9a71\u52a8\u66f4\u65b0\u3002 AMD\uff1a\u4ecefglrx\u9a71\u52a8\u8fc7\u6e21\u5230\u5f00\u6e90amdgpu\u9a71\u52a8\u7684\u8fc7\u7a0b\u4e2d\uff0c\u4f01\u4e1a\u7ea7GPU\uff08\u5c24\u5176\u6df7\u5408\u591aGPU\u67b6\u6784\uff09\u51fa\u73b0\u4e86\u517c\u5bb9\u6027\u6311\u6218\u3002 Intel\uff1a\u96c6\u6210GPU\u901a\u5e38\u4f9d\u8d56\u5185\u6838\u6a21\u5f0f\u8bbe\u7f6e\u9a71\u52a8\uff08KMS\uff09\uff0c\u521d\u59cb\u5316\u65f6\u53ef\u80fd\u4e0e\u4e13\u6709\u72ec\u7acbGPU\u9a71\u52a8\u4ea7\u751f\u51b2\u7a81\u3002 2.3 \u8f6f\u4ef6\u4f9d\u8d56\u51b2\u7a81 Xorg\u670d\u52a1\u5668\u7248\u672c\u4e0d\u517c\u5bb9\uff1a\u4f8b\u5982NVIDIA\u9a71\u52a8\u7684\u90e8\u5206\u529f\u80fd\u9700Xorg 1.20\u53ca\u4ee5\u4e0a\u7248\u672c\u652f\u6301\uff0c\u5728\u65e7\u7248Xorg\u73af\u5883\u4e2d\u4f1a\u89e6\u53d1\u663e\u793a\u9519\u8bef\u3002 CUDA\/CuDNN\u7248\u672c\u5339\u914d\u95ee\u9898\uff1a\u6df1\u5ea6\u5b66\u4e60\u5de5\u4f5c\u8d1f\u8f7d\u5bf9\u7248\u672c\u4e00\u81f4\u6027\u8981\u6c42\u4e25\u683c\u2014\u2014\u4f7f\u7528CUDA 12.0\u4f46\u9a71\u52a8\u4ec5\u652f\u6301\u5230CUDA 11.8\u65f6\uff0c\u4f1a\u5bfc\u81f4\u8fd0\u884c\u65f6\u5931\u8d25\u3002 2.4 \u5bb9\u5668\u5316\u73af\u5883\u6311\u6218 Docker\/Kubernetes\u9a71\u52a8\u900f\u4f20\uff1a\u5bb9\u5668\u8fd0\u884c\u65f6\u65e0\u6cd5\u8bc6\u522bGPU\u8bbe\u5907\u7684\u60c5\u51b5\uff0c\u591a\u56e0\u7f3a\u5c11`nvidia-container-toolkit`\u6216cgroup\u914d\u7f6e\u4e0d\u5f53\u3002 \u865a\u62df\u5316\u51b2\u7a81\uff1aKVM\/QEMU\u4e2d\u7684GPU\u900f\u4f20\u9700\u56fa\u4ef6\u652f\u6301\u4e0e\u7cbe\u51c6\u7684PCI\u8bbe\u5907\u5206\u914d\uff0c\u8f7b\u5fae\u7684\u9a71\u52a8\u7248\u672c\u53d8\u66f4\u5c31\u53ef\u80fd\u5bfc\u81f4\u529f\u80fd\u5931\u6548\u3002 [\/vc_column_text][\/vc_column][\/vc_row][vc_row el_class=”blog-detail-section”][vc_column][vc_column_text css=””] 3. \u517c\u5bb9\u6027\u95ee\u9898\u7684\u56db\u6b65\u8bca\u65ad\u6d41\u7a0b \u6709\u6761\u7406\u7684\u68c0\u6d4b\u662f\u51c6\u786e\u5b9a\u4f4d\u95ee\u9898\u7684\u5173\u952e\uff0c\u9075\u5faa\u4ee5\u4e0b\u7ed3\u6784\u5316\u6d41\u7a0b\uff1a 3.1 \u83b7\u53d6\u786c\u4ef6\u4fe1\u606f \u901a\u8fc7\u7ec8\u7aef\u547d\u4ee4\u8bc6\u522bGPU\u578b\u53f7\uff1a lspci | grep -i vga # NVIDIA\u4e13\u5c5e\u4fe1\u606f\u67e5\u8be2\uff1anvidia-smi -L \u901a\u8fc7\u670d\u52a1\u5668\u7ba1\u7406\u9762\u677f\uff08\u5982Dell iDRAC\u3001HPE iLO\uff09\u4ea4\u53c9\u9a8c\u8bc1\uff0c\u786e\u8ba4\u7269\u7406GPU\u5b58\u5728\u72b6\u6001\u4e0e\u56fa\u4ef6\u7248\u672c\u3002 […]<\/p>\n
     Read More…<\/a><\/p>\n","protected":false},"author":1,"featured_media":26241,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[4816],"tags":[8488,8489,8490,8491],"class_list":["post-26245","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-america-dedicated-server-sc","tag-linux-gpu-driver-compatibility-sc","tag-server-gpu-optimization-sc","tag-nvidia-amd-driver-issues-sc","tag-kernel-driver-conflict-resolution-sc"],"acf":[],"yoast_head":"\n