The present study aims to assess and advance the prediction skill of solar radiation in high resolution weather forecasts by large-eddy simulations (LES). The GPU-Resident Atmospheric Simulation Platform (GRASP) was used to simulate the atmosphere around the Cabauw experimental site for atmospheric research (CESAR) in the Netherlands. Large-scale boundary conditions were provided by coupling the simulation to a general circulation model (GCM). Radiative tendencies were calculated using two di_erent implementations of the Rapid Radiative Transfer Method for GCMs (RRTM-G): one runs in advance of the simulation using pre-calculated atmospheric _elds, the other employs dynamically updated _elds during the simulation. Both con_gurations generated simulations of every day in 2016, which were compared to each other and validated using observations from the Baseline Surface Radiation Network (BSRN). This study revealed that the implementation of interactive radiation altered cloud representation in the simulations, in most cases causing clouds to rise. This process correlated with an increase in turbulence kinetic energy of up to 2 m2/s2 locally. The clouds that were raised tended to break up more often between 5 and 7 km altitude, leading to a decrease in average cloud fraction and increase in short-wave down-welling radiation. The results suggest that the implementation of interactive radiation enabled the development of cloud top entrainment instabilities, which could be responsible for the cloud breakup in these cases. Regardless of the chosen implementation of the radiative transfer method, large errors are made in the prediction of surface solar radiation. GRASP produced root mean squared errors (RMSE) of 122.4 W/m2 and 115.7 W/m2 using prescribed and interactive radiation, respectively, while the large-scale model used to provide the initial and boundary conditions to the simulation produced an RMSE of 87.3 W/m2. The large error in GRASP's prediction of surface solar radiation can partly be attributed to conversion errors made during GRASP's initialization of the thermodynamic state, which lead to erroneous diagnoses of the liquid water content in the atmosphere.