We are in the initial stages of development of a ''non-empirical'' numerical tool for jet-noise prediction in the airline industry, ultimately to treat complex nacelles and nozzles. The non-empirical demand leads to compressible large-eddy simulations, followed by post-processing to produce the far-field sound. Here we treat a simple cold jet with an axisymmetric geometry. The simulations leave out the subgrid-scale model (which causes too much dissipation on the present grid as is often the case in transitional flows), use slightly upwind-biased high-order differencing, and are preliminary in that a grid-refinement has not yet been performed. We do not use any unsteady forcing. The initial instability remains grid-sensitive, but the region with developed turbulence gives accurate statistics. The sound seen in the simulations is also realistic. The far-field sound calculations use the Ffowcs Williams-Hawkings (FWH) equation with a control surface that encloses the turbulence as much as possible, and the outside quadrupoles omitted. We focus on the influence of the surface location and the problem of closing the FWH surface at the outflow of the simulation. Though many physical and numerical issues are only partly resolved, the agreement with experiment is quite good for the soundΓs level, directivity, and spectral content.