A series of 40-day non-hydrostatic global simulations was run with the NASA Goddard Earth Observing System (GEOS) model with horizontal grid spacing ranging from 50 km to 3.5 km. Here we evaluate the diurnal cycle of precipitation and organized convection as a function of resolution. For validation we use the TRMM 3B42 and IMERG precipitation products and 4 km merged infrared brightness temperature, focusing on three regions: the contiguous United States (CONUS), the Maritime Continent, and Amazonia. We find that higher resolution has mixed impacts on the diurnal phase. Regions dominated by non-local propagating convection show the greatest improvement, with better representation of organized convective systems. Precipitation in regions dominated by local thermodynamic forcing tends to peak too early at high resolution. Diurnal amplitudes in all regions develop unrealistic small-scale variability at high resolution, while amplitudes tend to be underestimated at low resolution. The GEOS model uses the Grell-Freitas scale-aware convection scheme, which smoothly reduces parameterized deep convection with increasing resolution. We find that some parameterized convection is beneficial for the diurnal amplitude and phase even with a 3.5 km model grid, but only when throttled with the scale-aware approach. An additional 3.5 km experiment employing the GFDL microphysics scheme and higher vertical resolution shows further improvement in propagating convection, but an earlier rainfall peak in locally forced regions.