Actually the engine generates detail with finer resolution (up to ~1 inch) to get better lighting. Meshes for rendering are created from these data but they have reduced resolution; that doesn't matter much as the lighting compensates for it easily. So if you can see that fine resolution, it means that it was generated. Of course it depends on the speed, for normal use the faster you go the lower detail is generated, but it's adjustable.
Another factor is whether the data for tire simulation are going to be carried over to CPU, which would impose a lag of 2-3 frames. Meaning, it would need even more sophisticated prediction algorithm determining what to generate and fetch in advance.