In hydrodynamic and hydromagnetic turbulence simulations, high spatial and temporal accuracy is of the essence. This is why pseudo-spectral methods are to be preferred. However, because of their intrinsic nonlocality, such methods are not well suited for massively parallel machines. In this talk the Pencil Code will be discussed, which uses sixth order finite differences for various types of turbulence simulations at resolutions up to 1024 cubed meshpoints on up to 256 processors: forced turbulence, shear flow turbulence due to the magneto-rotational instability, and convection. For simulations in spheres the physically interesting domain is embedded in a box. Some important results will be discussed and also the issue of code maintenance, development by many people and code validation will be addressed.
- http://www.nordita.dk/~brandenb/talks/misc/PencilCode04.htm - Link to a previous presentation