QubesOS will support NPU? I have Framework Laptop 13 with Intel Core Ultra 7 165H. The battery life is not that good at all. I having hard time. Or should i wait for 4.3 version as out-the-box-experience?
1 Like
I guess it’s possible to pass the NPU to a qube exactly like you would pass a GPU.
dom0 has certainly no use for a NPU at the moment anyway, so at least the device is not tied to it.
2 Likes
Check lspci if you have NPU device.
There is guide somewhere here how to use second GPU for LLM - search (I’ve seen it yesterday when searched for IO-SRV)
1 Like
In my AMD 7840U Framework 13 (with the older NPU) and Qubes 4.2, you can do passthrough, but the Linux kernel drivers do not support it at this stage (and it is not clear they ever will). I am not sure what is the situation with the Intel one.
1 Like
It is saying that at early of 2026 version 4.3 will released as stable version.
1 Like
Qubes OS 4.3 won’t change anything with regard to NPU or pass through support. That’s the qube that will receive the device that should have support for the NPU.
1 Like
This is keep me doubtful to having Framework Laptop 13 with Inter Core Ultra 7 165H. Or should the Framework Laptop 16 with AMD or AMD ai version? I know that QubesOS is consume more RAM, and energies. But v4.3 can uses NPU as acting as GPU?
1 Like
While I cannot say what’s the situation with the Ryzen AI processor’s NPUs, I would not get an AMD to use the NPU in Qubes (passthrough seems to not be supported due to hardware constraints, and the virtualization solution available is kvm-based so it is not clear it will be applicable to Xen) until I see someone getting it to work. That will not change with 4.3, I am quite sure.
I cannot say anything about Intel though, as their hardware (and drivers) are different. But no Qubes version will be able to use the NPU as a GPU, if that’s what you would like to happen.
I thought QubesOS depending more in GPU or NPU.
No, in fact use of the GPU (not even talking about NPUs here) in QubesOS is a bit of a sticking point. With a single GPU, you currently cannot have hardware acceleration in qubes, you need a second one you can passthrough (there are some experimental ways around that, but they are not fully developed and are not trivial).
From what I have read in the forums, QubesOS 4.3 will not change significantly the GPU situation (I’ll be happy to be told otherwise :-)).
What is really important for QubesOS is to have lots of RAM and a good (fast and reliable) SSD.
$700?
It’s an ARM CPU laptop, not compatible with Qubes OS which only supports amd64 / X86_64 CPU architecture.
Yes, I know. It was just a teaser for the devs, hopefully 
1 Like
This is a Xen limitation, not really Qubes OS itself.
In general, arm64 support is pretty bad because boards can be really different from one to another. Laptops using arm64 on the market are unfortunately often using out of tree kernel code and you end up with updates support similar to what you have with cheap Android smartphones 
Yes. I would like to find why Xen was chosen over other hypervisors…
There were really only two options, Xen or KVM.
The reason for picking Xen was the smaller trusted computing base, which means a reduced attack surface.
https://www.qubes-os.org/attachment/doc/arch-spec-0.3.pdf
This article on the latest VM escape highlights the security difference between Xen and KVM, even if it’s somewhat Xen biased.
1 Like
I see. Although, it’s more about QEMU, and a specific, targeted attack to KVM… With this remark I’m not opposing or pledging for any, just noticing for this specific case