Looking Glass Questions & Alternatives

How would one use looking glass to switch between VMs (that are using GPU passthrough) if you want to use only 1 display and 1 keyboard? I would like to be able to switch between VMs using the keyboard without any noticeable latency.

I was thinking this would be more of an “extra hardware” application - where you plug in your GPUs outputs into another card, which has some OS software to switch between them and is connected to only one display. However, I have heard about SR-IOV, but haven’t researched it yet.

Thoughts on the optimal method for this?
Thanks!

Looking Glass does allow what you are asking for as a software solution, it behaves just as any other application once it is setup. Think VNC but with hardware acceleration.

For how to setup Looking Glass please see:

https://looking-glass.hostfission.com/wiki/index.php?title=Installation

Thanks. I’m hoping for a non-software solution. Does SR-IOV allow for 1 screen, 1 GPU per OS, with no latency + switching between VMs instantly via keyboard?

Most SR-IOV GPUs don’t have any outputs at all, they are designed to use a solution like Looking Glass to obtain the frame buffer. SR-IOV still requires a software solution.

DirectGMA on AMD’s enterprise GPUs is another non software solution that you could look into.

Ah so maybe a Pro Vega II could do what I’m hoping on Linux without closed drivers! :slight_smile:
Looking into it…

I’ll also be trying to get 144hz working in Looking Glass… Hoping I can get the full 144hz (or 165hz / 240hz) working depending on which display I get.

144hz @ 4k is unlikely to perform well under load (ie, in games), this is a known limit of the current capture technology.

Right… I’m wondering about latency not due to gfx processing… Just the desktop experience, and using apps like Adobe Premier in a guest OS using Looking Glass.

Provided the capture is fast, the latency of the transfer is in the microseconds. The upload into the 2nd GPU however may incur additional latency depending on the hardware implementation. In most setups you should not see more then a frame or two of latency at most.

With the new AMD RDNA 2 let you assign a GPU per VM with one of these features?