I think he said sound isn’t being worked on atm, so for now passing through a USB sound card is the best option.
There is another developer working on Qemu sound emulation, have a look on the VFIO subreddit for it. Pulse patches provided work really well, it’s what I am using.
Has anyone looked into making this work on Mac OS?
Great thanks I’ll have a look into that
Undoubtedly it wont take long for a well maintained AUR to pop up.
There is no reason why it can’t, if Qemu and VGA passthrough works on Mac then it will be just a matter of porting whatever is needed in the client application to make it work.
Another little update,
Over the last few days I have been working on some things to better sync the incoming frames to the host’s GPU.
This video is with VSync Disabled, so there is tearing, but you can see once it is past the frame tear if you look carefully you can see the client (left monitor) is slightly ahead of the physically attached monitor at times (hard to see due to LCD latency for black->white)
We are actually getting the frame from the host while the host is still waiting on vsync.
With vsync enabled I am now seeing a consistent average of around 1-10ms latency @ 60Hz
we are getting so close
You’re using Nouveau? WHAT IS THIS MADNESS!?! Nouveau doesn’t even have reclocking capability yet!
Umm, where did you get that from?
wendell used nouveau in this video
Nope, the VM guest has the GTX 1080Ti, unless all of a sudden Nouveau supports AMD
Edit… I rewatched, he has 2x nvidia in that build, my mistake
@gnif Hello, amazing work you got there.
I am wondering, if intel on host would work, because I use Intel igvt-g (sharing igpu between host and vm) that doesn’t support outputs, but KVMFR looks like answer.
As I understand if directx capture works it should work?
That may not have been the best choice there @wendell. But for a quick and dirty test you shall be forgiven.
Nouveau is really sluggish. So I guess great test for worst case scenario.
The application that runs on the guest supports both NvFBC (NVidia Frame Buffer Capture) and DXGI Desktop Duplication. For NvFBC to work you need a quadro or better though (Feel free to complain to NVidia that they wont let any software except for Steam and ShadowPlay use this API on general cards).
How I have done the development is with a hacked GTX680 running as a Quadro K5000:
0c:00.0 VGA compatible controller: NVIDIA Corporation GK104GL [Quadro K5000] (rev a1)
So sick of NVidia’s BS on this stuff.
Please read the thread, this has been brought up many times now.
Actually the client application is barely doing anything, you could get great results on any card post 2013, and acceptable results on any card post 2003.
Is there difference on latency between NVFBC and DXGI capture? Sorry if it has been asked before but couldn’t find it.
Did you hack 680 by changing resistors or software hack?
I’m not sure If I’m disturbed or impressed.
But yes. Looking forward to testing and contributing to the git repo once released.
NVIDIA is a lost cause. Apart from a few dead cards of generations past, I just don’t buy NVIDIA’s stuff. I’ll be passing an RX 480 into the VM for looking glass.