Assessing the tidal stream energy resource, its intermittency and likely environmental feedbacks due to energy extraction, relies on the ability to accurately represent kinetic losses in ocean models. Energy conversion has often been implemented in ocean models with enhanced turbine stress terms formulated using an array-averaging approach, rather than implementing extraction at device-scale. In depth-averaged models, an additional drag term in the momentum equations is usually applied. However, such array-averaging simulations neglect intra-array device wake interactions, providing unrealistic energy extraction dynamics. Any induced simulation error will increase with array size. For this study, an idealized channel is discretized at sub 10 m resolution, resolving individual device wake profiles of tidal turbines in the domain. Sensitivity analysis is conducted on the applied turbulence closure scheme, validating results against published data from empirical scaled turbine studies. We test the fine scale model performance of several mesh densities, which produce a centerline velocity wake deficit accuracy (R2) of 0.58–0.69 (RMSE = 7.16–8.28%) using a k-Ɛ turbulence closure scheme. Various array configurations at device scale are simulated and compared with an equivalent array-averaging approach by analyzing channel flux differential. Parametrization of array-averaging energy extraction techniques can misrepresent simulated energy transfer and removal. The potential peak error in channel flux exceeds 0.5% when the number of turbines nTECs ≈ 25 devices. This error exceeds 2% when simulating commercial-scale turbine array farms (i.e., >100 devices).