I have been keeping an eye on this. From my understanding this uses the Lattice Boltzmann Method, which is supposedly categorized as DNS. Therefor you need a very fine "Lattice" to resolve the smallest turbulence scales, as there is no other way to take turbulence into account. If they find a way to do sub grid modeling and nonuniform mesh, then I will be very interested.
The compute times they posted on twitter are absolutely insane compared to fvm, for a singe Titan XP card. The simulation on the cow was 215 million lattice points, 28k timesteps, 23 min compute, on one graphics card