It's just a question of having enough VRAM+RAM to fit the model into memory.