文章目录⚙️前提安装必要工具禁用 nouveau 开源驱动⚙️NVIDIA 显卡驱动安装在线安装显卡驱动离线安装显卡驱动⚙️下载驱动对应版本CUDA(看情况安装)离线安装⚙️AI支持容器支持配置容器运行时Docker 场景Containerd 场景K8s 常用4.4 验证容器 GPU 调用⚙️前提安装必要工具dnfinstall-ydnf-utils dnf-plugins-core kernel-devel kernel-headers gccmake注意需保证kernel-devel版本与当前运行内核完全一致。若版本不匹配请先执行dnf update kernel -y并重启系统。禁用 nouveau 开源驱动系统默认的 nouveau 开源驱动会与 NVIDIA 官方驱动冲突必须提前禁用# 写入黑名单配置echoblacklist nouveau|tee/etc/modprobe.d/blacklist-nouveau.confechooptions nouveau modeset0|tee-a/etc/modprobe.d/blacklist-nouveau.conf# 内核启动参数添加 nomodeset避免启动阶段加载驱动grubby --update-kernelALL--argsnomodeset# 重建内核镜像使黑名单生效dracut-f-v# 重启系统reboot重启后验证是否禁用成功无输出即为生效lsmod|grepnouveau⚙️NVIDIA 显卡驱动安装在线安装显卡驱动# 添加 NVIDIA CUDA 官方源dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo# 启用系统依赖仓库RHEL 8 系dnf config-manager --set-enabled powertools# RHEL 9 系该仓库已更名为 CRB执行dnf config-manager --set-enabled crb# 安装驱动并重启dnfinstall-ynvidia-driver nvidia-settings nvidia-driver-cudareboot离线安装显卡驱动找到NVIDIA驱动下载官网https://www.nvidia.cn/drivers/搜索对应版本驱动下载对应版本安装驱动启用 PowerTools 仓库sudodnf config-manager --set-enabled PowerTools使用dnf安装下载的驱动包dnf-yinstallnvidia-driver-local-repo-rhel8-595.58.03-1.0-1.x86_64.rpm安装NVIDIA驱动控制软件和cuda管理工具# 安装本地源配置dnf-yinstall./nvidia-driver-local-repo-rhel8-595.58.03-1.0-1.x86_64.rpm# 刷新缓存dnf clean all# 安装驱动dnfinstall-ynvidia-driver nvidia-settings nvidia-driver-cuda# 重启加载内核模块reboot查看能使用的cuda版本发现是CUDA Version: 13.2[rootlocalhost ~]# nvidia-smiThu Jul1616:30:472026-----------------------------------------------------------------------------------------|NVIDIA-SMI595.71.05 Driver Version:595.71.05 CUDA Version:13.2|---------------------------------------------------------------------------------------|GPU Name Persistence-M|Bus-Id Disp.A|Volatile Uncorr. ECC||Fan Temp Perf Pwr:Usage/Cap|Memory-Usage|GPU-Util Compute M.||||MIG M.||||0NVIDIA L20 Off|00000000:E1:00.0 Off|0||N/A 30C P8 26W / 350W|0MiB / 46068MiB|0% Default||||N/A|--------------------------------------------------------------------------------------- -----------------------------------------------------------------------------------------|Processes:||GPU GI CI PID Type Process name GPU Memory||ID ID Usage||||No running processes found|-----------------------------------------------------------------------------------------⚙️下载驱动对应版本CUDA(看情况安装)重要说明nvidia-smi中显示的CUDA Version是驱动支持的最高 CUDA 版本并非必须安装的版本。驱动向下兼容低版本 CUDA可根据业务需求选择对应版本无需严格对齐。离线安装最新版本下载https://developer.nvidia.com/cuda-downloads所有版本下载https://developer.nvidia.com/cuda-toolkit-archive这里根据下载的版本去安装我这里是AnolisOS8wgethttps://developer.download.nvidia.com/compute/cuda/13.2.1/local_installers/cuda-repo-rhel8-13-2-local-13.2.1_595.58.03-1.x86_64.rpmsudorpm-icuda-repo-rhel8-13-2-local-13.2.1_595.58.03-1.x86_64.rpmsudodnf clean allsudodnf-yinstallcuda-toolkit-13-2# 配置系统级环境变量对所有用户、服务均生效tee/etc/profile.d/cuda.shEOF export PATH/usr/local/cuda/bin:$PATH export LD_LIBRARY_PATH/usr/local/cuda/lib64:$LD_LIBRARY_PATH EOF# 立即生效source/etc/profile.d/cuda.sh# 验证驱动nvidia-smi# 验证 CUDAnvcc--version注意CUDA Toolkit 为纯用户态开发工具链不涉及内核模块安装后无需重建 initramfs、无需重启系统。⚙️AI支持容器支持NVIDIA Container Toolkit 安装配置用于让 Docker / Containerd 容器能够调用宿主机 GPU 资源。#官方源curl-s-Lhttps://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo|tee/etc/yum.repos.d/nvidia-container-toolkit.repo# 下载官方源文件并批量替换为中科大地址curl-s-Lhttps://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo|\seds#nvidia.github.io/libnvidia-container#mirrors.ustc.edu.cn/libnvidia-container#g|\seds#https://nvidia.github.io/libnvidia-container/gpgkey#https://mirrors.ustc.edu.cn/libnvidia-container/gpgkey#g|\tee/etc/yum.repos.d/nvidia-container-toolkit.repo# 安装工具包dnfinstall-ynvidia-container-toolkit配置容器运行时Docker 场景# 安装Dockersudoyum clean allsudoyuminstall-yyum-utils device-mapper-persistent-data lvm2sudoyum-config-manager --add-repo https://mirrors.tuna.tsinghua.edu.cn/docker-ce/linux/centos/docker-ce.reposudosed-is|https://download.docker.com|https://mirrors.tuna.tsinghua.edu.cn/docker-ce|g/etc/yum.repos.d/docker-ce.reposudoyum clean packagessudoyuminstall-ydocker-cesudosystemctlenable--nowdocker# 修改Docker镜像源sudotee/etc/docker/daemon.jsonEOF { registry-mirrors: [ https://docker.1ms.run, https://docker.xuanyuan.me ] } EOF# 重启docker服务systemctl daemon-reloadsystemctl restartdocker# 自动写入 Docker 运行时配置nvidia-ctk runtime configure--runtimedocker# 重启 Docker 生效systemctl restartdockerContainerd 场景K8s 常用nvidia-ctk runtime configure--runtimecontainerd systemctl restart containerd4.4 验证容器 GPU 调用# Docker 验证示例dockerrun--rm--gpusall nvidia/cuda:12.2.0-base-ubuntu22.04 nvidia-smi