We report the results of multi-scale modeling of ultra-narrow graphene nanoribbons (GNRs) that combines atomistic non-equilibrium Green's function (NEGF) approach with semiclassical mobility modeling. The variability of the transport gap and carrier mobility caused by random edge defects is analyzed. We find that the variability increases as the GNR width is downscaled and that even the minimum variation of the total mobility reaches more than 100% compared to average mobility in edge-defected nanoribbons. It is shown that scattering by optical phonons exhibits significantly more variability than the acoustic, line-edge roughness and Coulomb scattering mechanisms. The simulation results demonstrate that sub-5 nm-wide nanoribbons offer no improvement over conventional bulk semiconductors, however, GNRs are comparable with sub-7 nm-thick silicon-on-insulator devices in terms of mobility-bandgap trade-off characteristics. (C) 2015 Elsevier Ltd. All rights reserved.