This contribution discusses the challenges of implementing star formation and stellar feedback processes in galaxy simulations. Insufficient computational power and numerous poorly understood physical processes, force simulations to adopt sub-grid models of the interstellar medium. These may crucially bias results. We advocate for smaller (~ kiloparsec) scale simulations of the interstellar medium to guide the development of sub-grid models in larger simulations. In this vein, I show results on ever increasing scales ranging from ~1 h-1 kpc3 ISM simulations, to a 1 h-1 Mpc3 simulation of a galaxy forming at high redshift, to a larger cosmological volume, 6.25 h-1 Mpc3, evolved down to redshift 3. We find that galactic winds can be powered by SN implemented as point explosions in high redshift galaxies simulated with sufficient spatial resolution. Galaxies at lower redshift require alternative sub-grid feedback models and we present one possible solution to generate winds from galaxies at redshift ~4.