Tiled upscale slows down gen speed significantly, which should be disabled when there is enough vram

[giteeeeee]

Generating an image with 4 steps + 2 steps highre fix (*1.1 upscale):

Tiled Upscale kicks in regardless of 15g of free VRAM:

As you can see in the screenshot above, I have have 15g vram left after loading the model, but the tiled upscale kicks in regardless, taking most of the generation time of 7 seconds. Also there is this 'cleanup minimal inference memory' of 3 seconds. But the actual upscale only takes 1 second.

I understand these steps are for memory optimization. But it's obvious that a 1.1x upscale doesn't requre 15g of vram. So shouldn't there be an option to disable such optimization? Or to have a vram checking system before optimizations kicks in?

Or am I just using it wrong?