hey all, I would recommend creating two separate pipeline instances and moving them independently to the two separate gpus.