Estimated time frame for Windows guest support?

[GizmoChicken]

Hi Jike,

I've been hoping for better GPU virtualization on KVM for years, so let me express how pleased I am that you are working on this project. This is great news!

I notice that your related project, XenGT, currently offers experimental support for Windows guests. I wonder, has work begun to support Windows guests in KVMGT? If so, would you be willing to share any predictions on when we'll see similar support in KVMGT? I'm very eager to give it a try!

It seems that development of XenGT is progressing more rapidly than development of KVMGT. If the development of XenGT is progressing more rapidly because Xen is, in some way, more amenable to GPU virtualization, then by all means, I hope Intel will continue to focus on whichever hypervisor offers the quickest path to the best result. But on the other hand, if either hypervisor is equally amenable to GPU virtualization, then I hope that Intel will consider placing a higher priority on developing KVMGT.

Xen is great for many use cases. Indeed, if I were administering a farm of virtual servers, I'd probably go with Xen over KVM.

However, my guess is that KVM would be a better fit for the vast majority of those who want to make use of GPU virtualization with current platforms having Intel integrated graphics, including laptops. For example, in my case, I use a Linux distribution as the primary operating system on my laptop, but I occasionally need access to a Windows guest for running Windows-only software. For someone who only needs occasional access to a guest on a laptop, Xen just doesn't make sense, and KVM is the better option. (I also currenty use KVM/VFIO, and have previously experimented with Xen, to passthrough a high performance GPU to a Windows guest on a home desktop computer, but that's a totally different use case.)

Of course, if you're developing XenGT with an eye toward a future, yet-to-be-announced, high-end graphics solution (perhaps competing with Nvidia GRID), then by all means, focusing on XenGT makes sense. But I would wager that, in the meantime, KVMGT would be embraced by a much wider audience than would XenGT, including many laptop users, such as me, who only need occasional access to a Windows guest.

I look forward to testing KVMGT on a laptop once it offers support for Windows guests, and I truly hope that the above feedback is helpful.

Best regards,
GizmoChicken

[l1viathan]

Thanks for your interests, so-nice writings and so-good advice, heartly :)

At the very beginning, Intel initiated the graphics virtualization project by some request from business customer, the hypervisor used is Xen. So we launched the XenGT project.

Nowadays, we treat Xen and KVM equally. The graphics virtualization project, now with a code name Intel GVT-g, will support Xen and KVM simultaneously. As a matter of fact, from a developer's perspective, the core logic of graphics virtualization is Hypervisor independent. We are working on abstracting common MPT service interfaces into an ops, and letting vgt interact with that.

The abstraction is in progress but not trivial, may be or not be included in next release. After the abstraction, there won't be separate XenGT or KVMGT repositories - a single iGVT-g repository containing both Xen and KVM support will be created.

As you may know, KVMGT lacks Windows support mainly because that write-proection is missing. We will working on that once the abstraction is done. I can't give an ETA now but again, wirte protection is the only missing feature of KVMGT, we can add that in 1~2 months if things go correctly.

[GizmoChicken]

Hi Jike,

Thanks so much for your quick and thorough reply. Much appreciated!

Great news to hear that you are rapidly moving toward a single iGVT-g repository that will support both Xen and KVM. This must be a challenging task, so I won't take up any more of your valuable time with questions. Please just know that many of us are eager to test out GPU virtualization on KVM with Windows guests, and we have high hopes for your success.

Thanks again for your great work on the project!

Best regards,
GizmoChicken

[m-i-k-a-e-l]

Hi, what is the status now of VirtualBox(=QEmu/KVM?) support for a Windows guest?

[l1viathan]

VirtualBox ==QEmu/KVM?

Sorry I didn't get your point, VBox is yet another virtualization solution which is differrent from qemu/kvm?

PS, the release by end of Q3 will probably contain Windows guest support.

[m-i-k-a-e-l]

@jikesong

If I recall right, VirtualBox on Linux can use KVM as virtualization backend. If so, Windows guest support with video acceleration will bring that whole user experience to a new level not only in QEMU but also in VirtualBox.

Approximately what is your release date now for Windows guest support and video acceleration?

[ariscop]

It's my understanding that windows guests already work, at least partially. Have run windows 7 as a kvm guest with aero enabled, though only once.

Won't really be usable until patches are upstreamed and frontends start supporting it, particularly for keyboard/mouse input

[l1viathan]

Widnows guest works perfectly, please have a look at:

https://github.com/01org/KVMGT-qemu/wiki/declaration

[l1viathan]

By the way, the igvtg mailing list, https://lists.01.org/mailman/listinfo/igvt-g, will be the primary place for discussion and be constantly monitored.

[GizmoChicken]

GizmoChicken wrote on Feb 8, 2015:

Of course, if you're developing XenGT with an eye toward a future, yet-to-be-announced, high-end graphics solution (perhaps competing with Nvidia GRID), then by all means, focusing on XenGT makes sense.

AnandTech wrote on November 8, 2017:

I’m going to be surprised if Intel doesn’t develop a true top-to-bottom product stack that contains midrange GPUs as well – something in the vein of Polaris 10 and GP106 – but for the moment the discrete GPU aspect of Intel’s announcement is focused on high-end GPUs. And, given what we typically see in PC GPU release cycles, even if Intel does develop a complete product stack, I wouldn’t be too surprised if Intel’s first released GPU was a high-end GPU, as it’s clear this is where Intel needs to start first to best combat NVIDIA.

Hope this means that these "high-end" Intel GPUs will support iGVT-g. 👍