Engine manufacturers are continually committed to find proper technical solutions to meet the more and more stringent CO2 emission targets fixed worldwide. Many strategies have been already developed, or are currently under study, to attain the above objectives. A tendency is however emerging towards more innovative combustion concepts, able to efficiently burn lean or highly diluted mixtures. To this aim, the enhancement of turbulence intensity inside the combustion chamber has a great importance, contributing to improve the burning rate, increase the thermal efficiency, and also reduce the cyclic variability. It is well-known that turbulence production inside the combustion chamber is mainly achieved during the intake stroke. Moreover, it is strongly affected by the intake duct geometry and orientation with respect to a plane perpendicular to the cylinder axis. In this paper, different geometries of the intake duct are analyzed by means of a 3D CFD code, in order to foresee the flow evolution and the tumble motion development during intake and compression strokes. Tumble vortex collapse and turbulence production at the end of the compression stroke are analyzed in detail, since turbulence levels just before TDC have a strong impact on the burning rate. Analyses are carried out in motored conditions, with unsteady boundary conditions provided by a 1D code. Different intake pipe orientations and throat areas are considered, since these parameters control the intensity of the tumble motion during the intake stroke. In addition, different bore/stroke ratios and compression ratios are considered, which indeed affect tumble collapse and turbulence production at TDC. Some correlations are developed, helping to identify the optimal geometry, in terms of turbulence production, although some drawbacks may be found concerning the engine volumetric efficiency. 3D numerical results, finally, constitute the prerequisite information to build-up a 0D phenomenological turbulence model coupled to a quasi-dimensional combustion model, and embedded in a 1D modeling environment to predict burn rate and engine performance. The latter is presented in a companion paper.