At least in the recent past, a hindrance was that MacOS limited how much of that unified memory could be assigned as VRAM. Those who wanted to exceed the limit

At least in the recent past, a hindrance was that MacOS limited how much of that unified memory could be assigned as VRAM. Those who wanted to exceed the limits had to tinker with kernel settings. I wonder if that has changed or is about to change as Apple pivots their devices to better serve AI workflows as well.