What is the status of iGPU support for Qubes?

What is the status of integrated graphics (iGPU) support for Qubes? I was thinking about buying a new x86 laptop with a strong iGPU. When browsing the forum threads, it sounds like an iGPU can be applied only to one VM at a time (or to dom0), just as dedicated graphics cards. If so, this changes what I plan to buy.

1 Like

A GPU (integrated or not) can only be used by a single system at a time. So if dom0 is using the integrated GPU and there are no other GPU, you can’t pass a GPU to a qube.

2 Likes

I think dom0 is always using the iGPU, am I right? Considering the default QubesOS setup.

I think dom0 is always using the iGPU, am I right? Considering the default QubesOS setup.

I’ve checked it. Qubes seems don’t work with pci.hide_pci with integrated GPU, i think second GPU is required for passthrough

This depends what the primary display is in the bios maybe. If you don’t have an iGPU and multiple GPus, one is picked.

Consider a vanilla QubesOS setup on @novacustom NV41 hardware.

2 Likes

This thread has been quite informative! My assumption is dom0 having access to an iGPU wouldn’t improve system performance that much. I don’t plan to use dom0 for anything other than system administration. So all the apps that would benefit from a GPU are in VMs.

One option is to concentrate all graphics heavy applications into one VM that gets GPU access. Given how I want to separate my VMs, this isn’t that great of an option for me. The other option is to minimize iGPU spend, and just know some programs will run slow.

I like Novacustom in general, but I worry about Intel chips these days. I was thinking of picking up a new AMD based system instead. The rumor is their 395 Max chip will come out early 2025 (strong iGPU), along with their new X3D laptop chips (essentially no iGPU, but with extra cache). Given that I don’t want to switch which VM has access to a GPU all the time, I might just go with the X3D option. This is of course if I can find a manufacturer that sells it without a dedicated GPU (or with a removable GPU that I could sell to reduce the total laptop cost anyways). I guess it also depends on CPU core count for the X3D chips. If it’s comparable to the 395 Max, then that makes the decision pretty easy.

As an aside, I’m aware Qubes might not work on new hardware right when it comes out. But I’m confident it will work eventually. And I might be able to make a contribution to the hardware compatibility chart this way.

Just to be clear, an integrated GPU is a graphic card builtin the processor, it is required to display stuff on the screen. dom0 requires a graphic card to display your desktop, hence the graphic card can’t be shared to a qube.

Yes. The X3D chips have a very minimal iGPU, but they do have one.

There are systems without GPU, usually routers or embedded devices on which there is a serial console to get a “text display” to operate them.

Up! Some people on Telegram say iGPU virtualization for newer Intel chipsets is almost ready and will be included in 4.3 release, but there is not even a guide how to try it now!

Unfortunately, it seems Intel only want to support SR-IOV (the tech allowing to split a device into multiple virtual sub devices) for their big dedicated GPUs

Damn that sucks. So no GPUs for us :frowning:

Having SR-IOV on desktop but not mobile is still better than not having it at all. Especially if customers respond well. Intel is in a rough position as a company right now. Maybe they will decide to implement more chips with SR-IOV if that’s one of the few features they can win on in the short run.

As an aside, Level1Linux has a video about using SR-IOV on desktop already with the new Intel Pro GPUs https://www.youtube.com/watch?v=xii8bqmE6jk

Me who just bought desktop with a mobile CPU just because who cares

One user already has made SR-IOV work with a mobile CPU

1 Like

Yes, but I don’t think Intel officially supports SR-IOV on their mobile chips. If I’m wrong about this, I would love to know.

Many users are wary of using custom patches that aren’t officially supported to enable advanced features. Some people need QubesOS to be a stable daily driver. The risk reward doesn’t always pencil out when there is now a chance the computer breaks with an update.

It woule be much better if Intel officially supported SR-IOV on their mobile chips. That would lower the barrier to entry the most, potentialy bringing it in more users. More users hopefully grows demand for SR-IOV. More SR-IOV hopefully reduces friction for people to use QubesOS.

2 Likes

Linux 6.18 brings in SR-IOV support in the xe driver for Tiger Lake, Alder Lake, Raptor Lake, Panther Lake, and BattleMage Pro. Since Panther Lake isn’t out there’s a question as to whether it’s supposed to work for all SKUs or not. iirc 6.17 also supports Panther Lake and BattleMage Pro, but it’s not enabled by default.

Meteor Lake, Lunar Lake, and Arrow Lake will not get official SR-IOV support. Meteor Lake does have unofficial SR-IOV support in the i915 driver from the linux-intel-lts 6.12/linux branch.

1 Like

We definitely need this. Who cares if it is official or not, this is important.

Could be full of vulnerabilities or even a malware.

1 Like