We present an experimental comparison of tangible rotary knobs and touch-based virtual knobs in three output conditions: eyes-on, eyes-free, and peripheral. Twenty participants completed a simple rotation task on an interactive surface with four different input techniques (two tangibles and two virtual touch widgets) in the three output conditions, representing the distance from the locus of attention. We found that users were in average 20% faster using tangible knobs than using the virtual knobs. We found that tangible knobs retain performance even if they are not in the locus of attention of the users. We provide four recommendations for suitable choosing knobs based on tasks and design constraints.