[chao-]

Having gone this route in the past*, once I accepted that I needed a second GPU dedicated as owned by the VM, most issues went away. No SR-IOV battles, just the simplicity of "Linux owns GPU A, Windows VM owns GPU B".

If I am remembering my build from the time, I had an RX 580 for Linux, and a GTX 1070 as passthrough to the Windows VM. At first, it felt like I was "giving up" on solving some problem that felt like it should be solvable, but it worked so well that I couldn't argue with the results.

It also assumes you have a motherboard with the right slots and enough PCIe lanes to get the performance out of two GPUs, and assumes you have a PSU with the power budget to support both GPUs. It definitely was a compromise, not perfection.

*Approx 2017 to 2020, before Proton or when it was was still new/immature. I now no longer care enough about games to play one when it doesn't just work on Linux. I assume the author does not feel this way.

[yrjrjjrjjtjjr]

Wouldn't it make more sense to use igpu for host than two discrete ones?

[chao-]

Yes, that would work the same way! However, I was on an HEDT system, with no iGPU and many PCIe lanes to spare.