But its also a 800B sized model running on a ram constrained system with no GPU.
Techniques, GPUs, more ram, and faster disks can always speed it up. But the point being is they run on low end machines now. Its now an optimization problem, not a possibility assessment.