VirtualBox KVM Backend Adds Support For SR-IOV Graphics

[Mechanix]

I have used a VirtualBox .vdi with KVM for a month now - didn't even know it was possible until I more or less accidentally tried.

This is great news! Maybe finally my Windows drive will vanish from this PC. I only ever used it when running a small set of games didn't work well in KVM or with Proton.

EDIT: Not on consumer cards? @$€"%"#¤%!!!!!!

[dragorth]

Thank you. I have a 4090 but did not know that the hardware needed to support that.

So, technically, the consumer cards do support it, but Nvidia hides it from software. So the 20 and earlier series have some hacks to enable it.

[pabloski]

Before getting too excited:
1) It's for APUs only.
2) It's not mainline yet.

sadly they are trying to sell overpriced discrete graphics cards to gullible "enterprise" clients....however it is possible to flash the "enterprise graphics cards" vbios on consumer cards, effectively enabling SR-IOV

yes it is a hack, but it works....it remembers me the "resistor solution" to the geforce consumer cards, that magically became enterprise cards

[geerge]

Thank you. I have a 4090 but did not know that the hardware needed to support that.

You only spent a grand on a graphics card, be grateful nvidia isn't charging you per million frames rendered.

[darkbasic]

sadly they are trying to sell overpriced discrete graphics cards to gullible "enterprise" clients....however it is possible to flash the "enterprise graphics cards" vbios on consumer cards, effectively enabling SR-IOV

yes it is a hack, but it works....it remembers me the "resistor solution" to the geforce consumer cards, that magically became enterprise cards

Can you please pinpoint me to such procedure? Not that it would help since Intel decided to cripple Xe cutting out any media engine capabilities, but I'm curious. Thanks.

[darkdragon-001]

Maybe this implementation has virtual TRIM support to automatically free up space for sparse disk images without this 7 year old unfixed bug: https://www.virtualbox.org/ticket/16450.
more details at https://superuser.com/questions/6465...ommand-support.

[mppix]

Spice is still a slow protocol with a bunch of legacy crap around it, Yes, it's better then VNC, but it pales in comparison to something gtk or dbus output. Spice should be abandoned post haste in exchange for some other kind of VDI appropriate protocol. In the past when the patches were first merged I experimented briefly wiring up dbus to gstreamer with great results (keyboard and mouse were not wired up at this time) having a proper utility to wire it up to (for instance sunshine/moonlight) would be a far better solution then Spice is.

Agree. I was actually expecting it to become the "Linux remote desktop protocol" but I don't think we will see that now with Gnome, KDE etc wiring up RDP.

[Quackdoc]

Agree. I was actually expecting it to become the "Linux remote desktop protocol" but I don't think we will see that now with Gnome, KDE etc wiring up RDP.

RDP isn't bad, but it's still not ideal. Personally I would love to see a protocol based on something like nvidia's gamestream (Sunshine + Moonlight) or other lower latency solutions (despite being low latency, it is still quite robust.) Perhaps something webrtc based even.