How are you running these together, splitting the model somehow or did you mean different models on any one card at a time?