my guess is that a grid that's too coarse for DNS and no turbulence model won't find a solution at all, like running a 'laminar' case where turbulence is expected. both LES and RAS will model turbulence each in its own way. can't tell about the spectrum though...