The present study deals with the shedding process of the Kármán vortices at the trailing edge of a 2D hydrofoil at high Reynolds numbers. Investigations are performed in order to evaluate the ability of an unsteady numerical simulation to accurately reproduce the vortex shedding frequency. The vortex shedding frequency, derived from flow induced vibrations measurements, is found to follow the Strouhal law as long as none of the resonance frequencies of the hydrofoil is excited. For such lock-off conditions, the Kármán vortices exhibit a strong spanwise 3D instability. For Reynolds numbers ranging from 35’000 to 40’000, the torsion mode of the hydrofoil is excited with a substantial increase of the vibration level. In this case, the shedding frequency is locked onto the vibration frequency. The vortex roll-up, shedding and advection are well predicted by the computations as well as the 3D instability. Nevertheless, the numerical simulations underestimate the vortex shedding frequency of about 20%. No significant influence of the turbulence models, the incoming flow turbulence intensity and the mesh size is observed on the computed values. To investigate the issue of discrepancy between absolute values of measured and computed shedding frequency, we intend to address the sensitivity of the vortex shedding frequency to the boundary layer. It is indeed know that the boundary layer transition point to turbulence for this profile geometry is located slightly downstream of the midchord for an incidence angle of 0°. In contrast, the numerical simulation, as it solves the RANS equations, models the whole flow as being turbulent. Consequently, rough stripes are placed just behind the hydrofoil leading edge. For such a configuration, the boundary layer is turbulent from the leading edge in a similar way to the computations. Notable influence of the roughness is observed: The vortex shedding frequencies are significantly decreased in comparison to the smooth leading edge. The average Strouhal number St decreases from 0.24 to 0.18 and is nearly equal to the computational one, St=0.19. Kármán vortices being very sensitive to the boundary layer, differences in the transition points between experiments and computations have to be minimized in order to aspire good agreement.