We model deterministic broadband (0–7.5 Hz) ground motion from an 7.1 bilateral strike‐slip earthquake scenario with dynamic rupture propagation along a rough‐fault topography embedded in a medium including small‐scale velocity and density perturbations. Spectral accelerations (SAs) at periods 0.2–3 s and Arias intensity durations show a similar distance decay (at the level of 1–2 interevent standard deviations above the median) when compared to Next Generation Attenuation‐West2 (NGA)‐West2 ground‐motion prediction equations (GMPEs) using a power‐law exponent of 0.6–0.8 above 1 Hz in models with a minimum of . With a trade‐off from , the median ground motion is slightly increased by scattering from statistical models of small‐scale heterogeneity with standard deviation () of the perturbations at the lower end of the observed range (5%) but reduced by scattering attenuation at the upper end (10%) when using a realistic 3D background velocity model. The ground‐motion variability is strongly affected by the addition of small‐scale media heterogeneity, reducing otherwise large values of intraevent standard deviation closer to those of empirical observations. These simulations generally have intraevent standard deviations for SAs lower than the GMPEs for the modeled bandwidth, with an increasing trend with distance (most pronounced in low‐to‐moderate scattering media) near the level of observations at distances greater than 35 km from the fault. Durations for the models follow the same increasing trend with distance, in which produces the best match to GMPE values. We find that a 3D background‐velocity model reduces the pulse period into the expected range by breaking up coherent waves from directivity, generating a lognormal distribution of ground‐motion residuals. These results indicate that a strongly heterogeneous medium is needed to produce realistic deterministic broadband ground motions. Finally, the addition of a thin surficial layer with low, frequency‐independent in the model (with a minimum of ) controls the high‐frequency decay in energy, as measured by the parameter , that may be necessary to include as simulations continue to extend to higher frequencies.