setup for M/L (nvidia)

[Neal Becker <ndbecker2@xxxxxxxxx>]

Trying to setup for machine learning.  Here's what I tried.  Actually, we can just view dnf history.  In reverse chronological order:

1. install the nvidia device driver from rpmfusion:

sudo dnf install 

    61 | install /var/cudnn-local-repo-rhel7-8.3.2.44/libcudn | 2022-03-31 18:38 | Install        |    1  
    60 | install /home/nbecker/Downloads/cudnn-local-repo-rhe | 2022-03-31 18:37 | Install        |    1  
    59 | install cuda                                         | 2022-03-31 15:55 | Install        |   60  <
    58 | install xorg-x11-drv-nvidia-cuda                     | 2022-03-31 15:23 | Install        |    4 >
    57 | -y install --disablerepo=* /tmp/akmods.VU8cZxAa/resu | 2022-03-31 14:57 | Install        |    1  
    56 | install akmod-nvidia                                 | 2022-03-31 14:49 | Install        |   62 E<

After installing akmod-nvidia, we need to obtain cuda and cudnn direct from nvidia.  They have setup repos, which after installation

provide:

baseurl=https://developer.download.nvidia.com/compute/cuda/repos/fedora35/x86_64

and a local repo:

baseurl=file:///var/cudnn-local-repo-rhel7-8.3.2.44

While nvidia provides fedora specific cuda I could only find rhel7 version of cudnn.

The installations succeed and everything is working.  But with these new repos enabled update is broken:

 sudo dnf update
[sudo] password for nbecker:
Last metadata expiration check: 0:45:46 ago on Thu 31 Mar 2022 06:38:21 PM EDT.
Dependencies resolved.

 Problem 1: package nvidia-driver-3:510.47.03-1.fc35.x86_64 conflicts with xorg-x11-drv-nvidia provided by xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64
  - package nvidia-driver-3:510.47.03-1.fc35.x86_64 obsoletes xorg-x11-drv-nvidia provided by xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64
  - package kmod-nvidia-5.16.18-200.fc35.x86_64-3:510.60.02-1.fc35.x86_64 requires nvidia-kmod-common >= 3:510.60.02, but none of the providers can be installed
  - cannot install the best update candidate for package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64
  - problem with installed package kmod-nvidia-5.16.18-200.fc35.x86_64-3:510.60.02-1.fc35.x86_64
  - package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64 is filtered out by modular filtering
 Problem 2: package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64 requires libnvidia-glcore.so.510.60.02()(64bit), but none of the providers can be installed
  - package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64 requires libnvidia-tls.so.510.60.02()(64bit), but none of the providers can be installed
  - package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64 requires xorg-x11-drv-nvidia-libs(x86-64) = 3:510.60.02-1.fc35, but none of the providers can be installed
  - package nvidia-driver-libs-3:510.47.03-1.fc35.x86_64 conflicts with xorg-x11-drv-nvidia-libs provided by xorg-x11-drv-nvidia-libs-3:510.60.02-1.fc35.x86_64
  - package nvidia-driver-libs-3:510.47.03-1.fc35.x86_64 obsoletes xorg-x11-drv-nvidia-libs provided by xorg-x11-drv-nvidia-libs-3:510.60.02-1.fc35.x86_64
  - package nvidia-settings-3:510.60.02-1.fc35.x86_64 requires nvidia-kmod-common >= 3:510.60.02, but none of the providers can be installed
  - cannot install the best update candidate for package xorg-x11-drv-nvidia-libs-3:510.60.02-1.fc35.x86_64
  - problem with installed package nvidia-settings-3:510.60.02-1.fc35.x86_64
  - package xorg-x11-drv-nvidia-3:510.60.02-1.fc35.x86_64 is filtered out by modular filtering
===========================================================================================================
 Package                    Architecture   Version                      Repository                    Size
===========================================================================================================
Skipping packages with conflicts:
(add '--best --allowerasing' to command line to force their upgrade):
 nvidia-driver              x86_64         3:510.47.03-1.fc35           cuda-fedora35-x86_64          22 M
 nvidia-driver-libs         x86_64         3:510.47.03-1.fc35           cuda-fedora35-x86_64         168 M

If I read this correctly nvidia's version (which provides driver 510.47) obsoletes the one from rpmfusion (which is actually newer?).

The only solution I can think of for now is just disable those 2 nvidia repos after installation.

Has anyone setup M/L nvidia on fedora and have a better solution?

Thanks,

Neal

_______________________________________________
users mailing list -- users@xxxxxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to users-leave@xxxxxxxxxxxxxxxxxxxxxxx
Fedora Code of Conduct: https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives: https://lists.fedoraproject.org/archives/list/users@xxxxxxxxxxxxxxxxxxxxxxx
Do not reply to spam on the list, report it: https://pagure.io/fedora-infrastructure