I see. Thanks for the note. I think beyond 48GB of VRAM diminishing returns set in very quickly so I’ll likely stick to that limit. I wouldn’t want to use models hosted in the cloud so that’s out of the question.