So I have both an igpu and dgpu, which means this should be easy, yeah?
Namely, Ryzen 9 7900 and RTX 4070ti.
But for the love of god I can’t get any installation working correctly with this combo OOB.
I’ve tried Nix. I’ve tried Arch. I’ve tried Garuda. I’ve tried Endeavour. I’ve tried Pop! OS.
I’ve also tried Proxmox, Truenas Scale, and Unraid, Smart OS, and others.
But there’s always something wrong with display.
Dual GPU and Nvidia are both tricky on Linux, so I assume you’re being sarcastic :P.
Unfortunately I don’t have experience with it, but I know people have gotten it working
Sadly not sarcastic. Ideal is Radeon handling the base, and NVIDIA being used in passthrough.
They just refuse to cooperate.
What do you mean by passthrough here? Usually passthrough refers to passing through a GPU to a virtual machine. And there is no cooperation whatsoever required between the GPUs for that. That makes me think you’re talking about offloading: one GPU controlling the display, while the other does the heavy lifting of 3d rendering. Last time I checked - several years ago - that is impossible with the proprietary nvidia driver, unless you have hardware that supports that, like prime in laptops. The only way to do offloading to a nvidia card without such hardware was to use the open source driver nouveau. And at the time there was absolutely no point in offloading with nouveau because it had such terrible performance. Now, this might have changed on several fronts since then; so take it with a grain of salt.
Scenario 1. X11 “works”, wayland doesn’t. Trying to update NVIDIA drivers leads to boot failure.
Scenario 2. Wayland works. Only on igpu. Only via HDMI. Only on one monitor.
Scenario 3. Wayland works on Displayport. Doesn’t even recognize second monitor.
Scenario 4. Everything seems to work. Trying to do GPU passthrough fails.
Scenario 5. IGPU is hogging displayport, despite being connected via HDMI, thus preventing the DGPU passthrough on either HDMI or DP.
I don’t know what you are trying to do, but you could setup a gpu less Linux base System and then use a virtual machine to actually run your daily os. If this makes things easier
Could you elaborate a bit?
Isn’t Proxmox etc. “Gpu less”, as they only use tty instead of anything like a WM or DE?
I’d prefer a “master” / hypervisor running a bunch of VM’s for different purposes.
Whether they be for gaming, pirating, development, pen testing, home automation, porn, or anything else really.
'Course I’d only be running gpu passthrough into a single VM at a time, can’t split a single GPU into 50 passthroughs yet.
Well of course you can use proxmox or tuenas but for “more performance” I would just setup qemu to pass through gpus and input to the vm.
However I don’t know if this is worth the hassle.
It sounds like you’re trying to plug in multiple monitors across different GPUs…that’s a recipe for disaster. Plug all monitors into a single GPU. If you’ve got a 4070, you’re already more than capable of 4+ monitors on that card, what are you doing exactly?
Only good reason for doing what you’re doing is for something like GPU passthrough in proxmox where you’re passing your DGPU through to a virtual machine.
The integrated GPU in the 7900 is exceptionally weak, and generally only good for “home office” use.
How would hooking up everything to the GPU be beneficial when it comes to GPU passthrough?
Albeit is it even necessary these days.
If you’re doing GPU passthrough, then you’d have a dedicated monitor on the iGPU just for managing VMs, and you’d have a second monitor hooked up to the dGPU and it would initialize with the VM bootup. If what you’re trying to do is just get multiple monitors on a linux install, then you should just disable the iGPU and don’t plug anything into it at all, and only use nVidia driver stack for your setup.
I don’t know what problem you’re trying to solve exactly. Are you looking to do GPU passthrough on VMs, or are you just trying to get a standard multi-monitor setup on a gaming PC? What, exactly?
iGPU shares one monitor with the dGPU, but on different protocol, which from what I read online is supported.
It only really needs output when I flick it open.
So maybe it needs a KVM switch instead of trusting the monitors splits.
It’s possible you need a dedicated KVM for swapping the input of the iGPU/dGPU shared monitor. If the dGPU is only getting passed through to a VM, I don’t believe it will initialize until the VM explicitly takes control of it.
And I think vGPU passthrough is disabled on consumer Ada/Ampere architectures unless there’s been a crack developed for it since then?
The turing architecture (RTX 20 series) is the only one I know of that has a crack/drivers available for it but I may be wrong/out of date on that info.
It’s definitely an unusual setup but not that much unusual.
Not sure if it helps but I also had trouble with my Dell XPS and I tried a bunch of distros and their variants, and the only one that worked flawlessly out of the box was Fedora. And not the KDE version, GNOME was better. Give it a try, it’s really a great distro all by itself too.
Are you using the iGPU? Like, can you get away with disabling it in BIOS?
You can disable it explicitly, yes.
It should be possible to use it with the dgpu.
Edit: You can also prioritize using the iGPU over the dGPU in bios. Maybe that’d work, hmm.
