The paper outlines the predictive capabilities of lattice Boltzmann methods (LBM) in turbulent shear flows. Attention is devoted to a specific collision operator which relaxes the distribution functions in cumulant space. The study highlights the benefits of a carefully defined discrete collision operator by scrutinizing the numerical stability and the predictive accuracy for a wide scope of resolutions—ranging from DNS to RANS—when no ad hoc turbulence closure is employed. Examples included are concerned with two frequently computed fundamental flows, i.e. Taylor-Green vortex and channel flows. Results reveal a fair accuracy and a remarkably small resolution dependence for the investigated cumulant collision operator, which is quite the contrary for other collision models.