>>106119759
How come it's still that fast even though it offloads to RAM?
The last time I tried loading a model that couldn't fit into VRAM iteration time would go up a crazy amount.