There are several issues, though the primary issue is that after using Fedora 39 for minutes or hours, it locks up. I have detailed the upgrade process and issues in this blog post, thus will only summarize here.
When using monitors on both nVidia GeForce GTX 550 TI and AMD 6500 XT cards, it locks up (freezes) after about 10-20 minutes of use using default drivers (nouveau and AMDGPU).
When I disconnect the monitors from the nVidia card, and only use 1 monitor on the AMDGPU card, I can go a lot longer before freezing, but it eventually freezes. In this scenario, it seems to only freeze when interacting (moving the mouse), and not typically while a video is just playing.
AMD video only works with X11. When I use Wayland, you can see the logo at bootup, but after that, it is black, though you can always see the mouse cursor when you move it across it. This is true for all 3 monitors I tested, which work just fine with nVidia (native and nouveau).
As noted in the blog post
This PC is rock solid with Fedora 28 and native nVidia drivers, and NEVER locks up. Because the AMD card was released in 2022, Fedora 28 with kernel 4.9 cannot use it.
Using native nVidia drivers on Fedora 39 is not an option, and the primary goal here is to migrate over to just using AMDGPU, though it would be a plus if I could use both cards indefinitely with open source drivers.
While getting the AMD card to work with Wayland instead of X11 would certainly be very nice to have, it is not currently a show stopper. The computer locking up is a show stopper.
It looks like that nvidia GTX 550 Ti GPU is only fully supported by the nvidia 390xx driver which can be installed from rpmfusion-nonfree. That repo can be enabled as shown here and the driver would then be installed as shown here.
Note that for that legacy nvidia card the only DE that can be used is X11. The driver does not support wayland.
It is possible the problem causing the freeze is the extreme difference between the hardware – geforce legacy circa 2011 and AMD circa 2022. Even with the suitable drivers that is a severe difference in hardware. If you could simply remove the nvidia GPU it likely would make a difference.
This also may be be related to the age of the overall machine (mobo and bios) as well, assuming it is aged as is the nvidia GPU and has a relatively new AMD GPU installed.
Thank you. I’ll experiment and report back. The problem with removing the nVidia card – which I am OK doing eventually – is that I need to be able to fall back on F28 until I have everything running on the new. That last 1% could take months. Having to swap video cards is not an ideal way to do that. I didn’t see any option to disable a PCIe slot from the BIOS, and certainly can’t do for the #1 slot, as that is the only one that BIOS will post with.
The MOBO, CPU and RAM is newer than the nVidia cards, compliments of Corsair when their water cooling leaked. Thus, I’m still a fan of Corsair. Zero problems with their newer coolers to-date.
The brand new M.2, released 2022, is working great in it. This is a gen3 PCIe.
As for installing drivers via non-free, been a very long time, but don’t remember them stopping you from installing drivers that just don’t work cuz nVidia stopped supporting. Thus, I do remember pulling my hair out not understanding why they didn’t work with the Fedora 28 upgrade until I read on nVidia’s own website that they quit supporting that card as of kernel 5.0. Thus, I dread the thought of even trying to taint the OS with those drivers when the primary goal is to be free of the native dependency. I can live with only using the nVidia card when I’m on Fedora 28, which is ultimately temporary.
Perhaps when booting with only the ADMGPU monitor connected, I can try blacklisting nouveau to further take it out of the loop. That’s easy to do with a grub edit during boot, and I can even copy the grub entry long-term if that helps.
UPDATE: Took nVidia out and moved AMD to slot #1. Been running dual screen with no issue so far for 18 hours. Though, ran all night Friday night with just nVidia monitors unplugged before lockup, so will take awhile to have confidence.
Lockups may be a pre-fail indicator with old hardware, so it is best if you can work without the old Nvidia card – quoting rpmfusion.org HowTo:
Legacy GeForce 8/9/200/300
Supported on current stable Xorg server release. EOL by NVIDIA at the end of 2019. Still available on “best effort basis” (newer kernel may break, will be discontinued at anytime if not actively maintained).
Except that I have never had a lockup with the nVidia on Fedora 28, and I reboot only once a month and use it from morning to night every day as my primary workstation. For those of us who build our own future proof builds with quality parts, it’s not “old” until about 20 years. None of my nVidia cards have shown any sign of failing.
I think the primary issue is simply with the nouveau drivers, which I do not use with Fedora 28, but do use with Fedora 39. When I blacklisted nouveau with nVidia in slot #1 (2 monitors), AMD in slot #3 (1 monitor), I got a low res single monitor display, so couldn’t even access the AMD, making this an unviable test scenario.
Once I’m confident that the AMD is working great by itself (so far so good), I plan to try the nVidia in slot #3, and see if I cannot boot into F28 with that. In theory, I can blacklist nouveau in F39, but this time with dual monitor working on the AMD. Hoping that is the solution.
I have overall very good news to update with. I put the AMD GPU in by itself and ran 40 hours without lockup. I then added the NVIDIA card in slot #3, and again ran 40 hours without lockup. During these times I stressed it with things like phoronix test suite, gaming, tons of multi-tasking and watching videos.
This vindicates the nouveau drivers. This also means that I can indefinitely run 3-4 1080p monitors now on my F39 install, which is a huge icing on the cake. Because this is in my living room, I do have plans on having 3 monitors (43=60") vertical from my ceiling on down, plus a 27" to the right.
The conclusion is that it came down to balancing the cards with the PCIe slots, where the MOBO seems to want the AMD GPU in slot #1. Here is a breakdown to better understand it:
The MOBO has 4 gen3 PCIe slots. Only slot #1 is 16x. The rest are 8x.
The NVIDIA card (GeForce GTX 550 Ti) is a gen2 16x.
The AMD card (Radeon RX 6500 XT) is a gen4 4x.
In benchmark testing, the AMD card is nearly 5x faster than the NVIDIA card.
And it is this last point that I believe really matters, because faster translates to increased PCIe bandwidth.
This configuration results in the perfect end-state for the new Fedora 39 install. The only downside is that I’m unlikely to boot into the GUI desktop of F28 (or my F16 install lol) as that requires putting the NVIDIA card back in slot #1. But, I can still boot into those older installs in text mode by editing the grub entry on boot and inserting single. Doing this, I can, for instance, run my backup scripts one last time. But, I can’t do any exporting or backup from any GUI tools like web browsers. I can live with that temporary restriction.
I could share more on the overall contrast from the upgrade, such as LVM being backward, but not forward compatible. I think that can be better placed in a new post, though.
Note that I have not tried switching back to Wayland. Not in a hurry to try, as x.org is currently working well.
