Red Hat Is Rolling Out A VirtIO DRM/KMS GPU Driver

[phoronix]

Phoronix: Red Hat Is Rolling Out A VirtIO DRM/KMS GPU Driver

Red Hat has published a new KMS driver for the VirtIO GPU used within their Linux virtualization stack...

http://www.phoronix.com/scan.php?page=news_item&px=VirtIO-DRM-KMS-GPU-Driver

[shmerl]

Does it mean KVM virtual machines will get a good 3D acceleration?

[geearf]

Does it mean KVM virtual machines will get a good 3D acceleration?

I believe that's the goal of Virgil3D.

[degasus]

Does it mean KVM virtual machines will get a good 3D acceleration?

Depends on what you mean with "good". It's likely neither fast nor feature complete with the host, but it sounds like a nice way to get a portable and generic 3d acceleration at all.

[chrisb]

Depends on what you mean with "good". It's likely neither fast nor feature complete with the host, but it sounds like a nice way to get a portable and generic 3d acceleration at all.

It shouldn't be that slow. Intel benchmarked the overhead of kvmgt at somewhere between 5% and 15%. That might matter for gamers, but most desktop users wouldn't notice the difference.

[elmarco]

vs

It shouldn't be that slow. Intel benchmarked the overhead of kvmgt at somewhere between 5% and 15%. That might matter for gamers, but most desktop users wouldn't notice the difference.

As usual, it's difficult to compare things, because there are various differences and trade-offs. virgl is designed to work with any opengl 3.3 stack, and doesn't rely on hw extensions. The memory isn't shared between host and guest either, and that's one of the reasons for slowdown compared to kvmgt. But that depends a lot on how opengl is being used. For now, it's too early to compare the two, but it's likely that kvmgt will be faster in general. I'd also keep in mind that today kvmgt benchmark are run with a dedicated output, the result is not yet composed on a client desktop.