Well… I guess that probably depends on the context and how you interpret the word “optional”. 
When I said optional, I meant it’s a package that isn’t one of the packages akmod-nvidia requires and automatically installs. And the cuda package isn’t required for the driver itself to work. And that was (incorrectly) my way of thinking when I went to do the upgrade. Again, I freely admit I did the upgrade wrong based on the packages I had installed.
You’re correct, once cuda is installed it’s no longer optional with respect to the version of the drivers that are installed.
I’m assuming if I had done the upgrade like either one of these approaches it would have worked fine.
- Explicitly add it on the upgrade command:
# dnf upgrade nvidia-550.90.07/*.rpm nvidia-550.90.07/optional/xorg-x11-drv-nvidia-cuda-550.90.07-1.fc40.x86_64.rpm
- -OR- First remove cuda, then do the upgrade
# dnf remove xorg-x11-drv-nvidia-cuda
# dnf upgrade nvidia-550.90.07/*.rpm
The cuda package could then be “optionally” installed at any later point if desired, it wasn’t actually needed to test the driver. In fact, the only reason I installed the cuda package is because that’s where the nvidia-smi command lives and it’s a useful command for verifying both the version of the current driver and what processes are using the driver… and I’m sure many other things.
I can’t remember if I tried that or not. I maybe did, but can’t be sure. And I don’t remember the results if I did. I’m not ready right now to upgrade to 555 or 560 in order to test it. If I had a separate system dedicated to testing I’d be more inclined, but this is my main laptop and the only one with an nvidia GPU, so I’m leaning towards leaving the driver as is for now. If I get some spare time and the inclination, I might be willing to test that and try to narrow down exactly which 555/560 release I start to see the “freezing on external display” issue.
I did see some of those reports from others having freezing issues, but it seems like many of them were different from my experience. For me, it was only when I ran windows-based games (ie games needing Proton) on the external display. While I had 560 installed, I didn’t notice freezing issues any other time. Not saying it’s necessarily a different cause, they certainly could all be related, just saying I don’t have any direct experience with other freezing issues.
Which made me wonder a bit why the rpmfusion-nonfree-updates repo moved to the 555/560 drivers. When I first installed the F40 nvidia drivers (8/21), they were at 555.58.02, and I just assumed (incorrectly) that was a “stable” release. It wasn’t until I starting having problems and looked at the NVIDIA driver page that I realized 555 was the “new feature branch”.
I get rpmfusion is more bleeding edge than the normal fedora repos, but since it’s the only “repo-ish” way to get the nvidia drivers for F40 it kind of feels like 555/560 would have been more appropriate in the rpmfusion-nonfree-updates-testing repo… or maybe some other repo. Admittedly, I have very little experience in this area, so completely understand I may be missing the mark. Just seems like it makes it more complicated for someone to get the “recommended/certified” version of the nvidia drivers.
From what I understand there’s also a cuda repo apparently managed directly by nvidia? But it hasn’t been updated for F40, only F39. It also seems a little more involved than the rpmfusion method… at least based on the link, I have no direct experience with it.
And, there’s the manual driver installation method by getting the drivers direct from NVIDIA, but that method just seemed more complicated to setup and manage than I wanted.
If 550.107.02 for F40 becomes available on koji.rpmfusion, I’d very likely be willing to give it a try. There actually is a build for 550.107.02, but it was cancelled, so the rpms don’t currently exist.
