NVIDIA drivers and toolkits are used across many different models of GPUs.
This all worked up until I updated Bluefin to the Fedora 42 version.
So the update is the likely culprit (either the update has some issues, or I messed up the upgrade somehow?
Anyway, I’m probably going to move over to the GDX version this week.
If anyone had told me in 2021 that I would have been running Centos on my desktop in 2025 I would have told them “Sweet, that’s what I’m running for work now anyway & it’s stable as all F!”
+1 on CentOS. I ran that for years in AWS EC2 servers.
Be aware that there is a lot of churn right now with the nvidia drivers. I am not certain because I do not have a 4070 device, but I have a hunch it might be better supported with the open source driver now. People’s experience seem to be all over the map.
My 3050 is working with the proprietary driver still and so I have not dug deep yet.
Anyone who can't wait for an update, the following will:
#add a new flatpak repo and install a fixed version of Obsidian
# Not recommending you should do this, but if you need Obsidian then either, wait, manually start it with the --diable-gpu flag or do it this way.
flatpak install --user https://dl.flathub.org/build-repo/186356/md.obsidian.Obsidian.flatpakref
#Test run if you want.
flatpak run --user md.obsidian.Obsidian//test
Otherwise, just click on the launcher and it should be working.
After doing the upgrade, it works fine now
I can’t recall the last time I had an Nvidia-related problem. Using Bluefin, the drivers have switched from proprietary to open, and I’m on GTS now, so I assume back to proprietary. No problems.
My GPU is just for cuda, and doesn’t drive my displays.
I am right there with you @JohnAlt. But I had so many weird problems on Fedora WS because of bad (outdated?) advice - I am spooked. And nvidia drivers are in transition again.
I don’t trust the combination of linux kernel, nvidia drivers and correct process being followed to make them work together. But, CUDA … so I am living with it.
When @NoYards reported this I just assumed that I probably was impacted but just hadn’t felt it yet.
BTW, I finally carved some time to test CUDA in bluefin-dx-nvidia-open - works like a charm. I tested on the host using the ‘Test CUDA support’ snippet from ujust configure-nvidia, pytorch samples I wrote on the host, as well as some examples using cupy/cupy published OCI image. I just had to tweak my usage of podman in the last case - it was out of date.
If interested, I pushed my podman usage tweak in the run_cupy script here.
Thanks again for testing that stuff during the beta. That is the only reason I took the time now to try it.
