2 ms·

I came here wondering the same. Is this based on work done by folding@home for instance? (As in, it used their precomputed stuff as training data)