Seeking a better understanding

Operating Systems & Open Source Looking Glass
1

I just recently stumbled onto this project and it sounds amazing! Can somebody fill in these blanks for me:

From the LG website:

allows the use of a KVM configured for VGA PCI Pass-through without an attached physical monitor

From what I’ve gleaned, it seems to me LG is independent of the hypervisor and relies only on the Linux frame buffer client and Windows frame buffer server.

Am I wrong? If so, what is LG’s dependency on KVM?

What frame buffer capture methods are currently supported?
How does LG decide which method to use?
Based on experience, what are the practical differences?

GPU Capture API Max viable resolution
nVidia GTX DXGI 1980x1200
nVidia Quadro NvFBC ~4k
AMD Radeon DXGI ?
AMD Radeon Pro DOPP ?

This video mentions a headless host which employs two dGPUs, each being passed through to a different guest.

Is there any reason I couldn’t do the same with an Intel iGPU + one Quadro like so:

System(s) OS GPU Interaction
Host Linux Headless SSH
Primary Guest Linux Intel iGPU, configured for mediated PCIe passthrough (SR-IOV a.k.a Intel GVT-g) Monitor
Secondary Guests Multiple Shared Intel iGPU Spice
CAD Guest Windows Quadro Looking Glass
Gaming Guest Windows Quadro Looking Glass
2
  1. Almost, it still uses QEMU to virtualize the device to share RAM and if you’re using SPICE for keyboard & mouse input.
  2. DOPP is not implemented, and on modern versions of LG (B1) you can attain 4K with DXGI, however your results may vary depending on your workload. The only practical difference between the methods is performance, and this can vary wildly depending on your system.
  3. As long as you can pass a GPU into each it will work, however please note that some of the Intel iGPUs suffer texture upload performance issues, and as such the performance may be lacking.
1 Like
3

Thank you.

How does LG select the API it uses to access the GPUs frame buffer? Is it as simple as Quadro = NvFBC and for everything else it’s DXGI?

My dGPU is a modern Quadro. Full screen resolution is 3440x1440. Is there any way to reliably determine which Intel iGPUs will suffice or which would cause issues due to a lack of texture upload performance?

4

LG simply tries NvFBC and if it fails, it reverts to DXGI. It will show this in the host application log.

No I am sorry. 4K resolutions for LG are already at the upper limit for capture, adding to that a slower IGP you’re likely to see low update rates.

1 Like