Do all LES use finite volume methods?

No

WHat are the advantages of using finite volume over finite difference for LES?

- Finite difference can use more accurate schemes (for spatial discretization). For LES this advantage is very important for academics research. If you can reduce significantly numerical error cause by discretization , you can most understand the effect of LES model.

-Finite volume can shape more sophisticated body. It's a very important advantage in the case of industrial research.

In the two case, you can use LES. But like everything, there are 'no magic solution'. In the two cases, you pay a price in order to have something.

How long would it take to convert a RANS solver to a basic LES solver?

I don't know.

The difference is really between structured and unstructured grids. You can have very accurate finite volume schemes on structured grids but the same is not possible with unstructured grids.

You can, indeed, but at a very high cost. Furthermore, a high order shape function to reconstruct the fluxes and a high order surface integration impact the structure of the matrix and thus the robustness of the numerical method.

You can't have the same accurate with finite volume. For example, you can't try a finite volume with the same accurate and cost than high order ompact schemes. And when you want to study some LES model or very special turbulence cases, finite volume is not the right way !

But you're right, -the- difference is really between structured and unstructured grids. But, really can you can explain that is the difference between finite volume and finite element in the case ?