Looks like it basically gives the focused application highest priority for VRAM allocations and deallocations, which lower VRAM devices would be doing more often.

I’m kinda surprised this wasn’t already the default behavior, as the application the user has focused should almost always be given render priority, regardless of what it is.