The recent passing of the petaﬂop/s landmark by the Roadrunner project at the Los Alamos National Laboratory marks the preliminary peak of an impressive world-wide development in the high-performance scientiﬁc computing sector. Also purely academic state-of-the-art supercomputers as the IBM Blue Gene/P at Forschungszentrum Jülich allow nowadays to investigate large systems of the order of 1000 spin polarized transition metal atoms by means of density functional theory. Three applications will be presented where large scale ab initio calculations contribute to the understanding of key properties emerging from a close interrelation between structure and magnetism. The ﬁrst two examples discuss the size dependent evolution of equilibrium structural motifs in elementary iron and binary Fe-Pt and Co-Pt transition metal nanoparticles, which are currently discussed as promising candidates for ultra-high-density magnetic data-storage media. However, the preference for multiply twinned morphologies at smaller cluster sizes counteracts the formation of a single-crystalline L10 phase which alone provides the required hard magnetic properties. The third application is concerned with the magnetic shape memory eﬀect in the Ni-Mn-Ga Heusler alloy, which is a technologically relevant candidate for magneto-mechanical actuators and sensors. In this material strains of up to 10% can be induced by external magnetic ﬁelds by the ﬁeld-induced shifting of martensitic twin boundaries, requiring an extremely high mobility of the martensitic twin boundaries, but also the selection of the appropriate martensitic structure from the rich phase diagram.