我正在尝试使用此存储库安装特定版本的 nvidia 驱动程序: https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rehl8.repo
在运行的 docker 容器中
nvidia/cuda:12.2.2-cudnn8-runtime-ubi8
repo 的索引显示我想要的 RPM 版本就在那里,
nvidia-driver-525.85.12-1.el8.x86_64.rpm 43MB 2023-01-28 17:37
nvidia-driver-525.105.17-1.el8.x86_64.rpm 43MB 2023-03-29 08:56
nvidia-driver-525.125.06-1.el8.x86_64.rpm 43MB 2023-06-02 03:03
nvidia-driver-525.147.05-1.el8.x86_64.rpm 43MB 2023-10-26 06:12
nvidia-driver-530.30.02-1.el8.x86_64.rpm 42MB 2023-02-23 05:25
nvidia-driver-535.54.03-1.el8.x86_64.rpm 165MB 2023-06-13 00:04
nvidia-driver-535.86.10-1.el8.x86_64.rpm 158MB 2023-07-27 06:46
nvidia-driver-535.104.05-1.el8.x86_64.rpm 158MB 2023-08-21 01:23
nvidia-driver-535.104.12-1.el8.x86_64.rpm 158MB 2023-09-22 05:42
nvidia-driver-535.129.03-1.el8.x86_64.rpm 158MB 2023-10-20 04:20
nvidia-driver-545.23.06-1.el8.x86_64.rpm
但是当我添加存储库时:
dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo
并列出可用的软件包
$ dnf list --showduplicates | grep nvidia-driver
我得到了输出
sh-4.4# dnf list --showduplicates | grep nvidia-driver
nvidia-driver.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-NVML.i686 3:545.23.06-1.el8 cuda
nvidia-driver-NVML.i686 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-NVML.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-NVML.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-NvFBCOpenGL.i686 3:545.23.06-1.el8 cuda
nvidia-driver-NvFBCOpenGL.i686 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-NvFBCOpenGL.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-NvFBCOpenGL.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-cuda.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-cuda.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-cuda-libs.i686 3:545.23.06-1.el8 cuda
nvidia-driver-cuda-libs.i686 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-cuda-libs.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-cuda-libs.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-devel.i686 3:545.23.06-1.el8 cuda
nvidia-driver-devel.i686 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-devel.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-devel.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-libs.i686 3:545.23.06-1.el8 cuda
nvidia-driver-libs.i686 3:545.23.06-1.el8 cuda-rhel8-x86_64
nvidia-driver-libs.x86_64 3:545.23.06-1.el8 cuda
nvidia-driver-libs.x86_64 3:545.23.06-1.el8 cuda-rhel8-x86_64
都不是我想要的版本。
尝试安装我想要的版本也会导致找不到软件包错误。
我是否遗漏了某种隐藏逻辑,为什么我只能安装最新的软件包?
编辑:我还应该补充一点,我可以安装最新版本
545.23.06-1
,但它找不到低于该版本的任何版本
看起来正确的答案是容器不应该安装 nvidia 驱动程序。相反,他们应该安装使用主机级 nvidia 驱动程序的 cuda-toolkit。
我将更新图像以使用 cuda-toolkit