A computational framework is presented for describing the nucleation and growth of vacancy clusters in crystalline silicon. The overall approach is based on a parametrically consistent comparison between two representations of the process in order to provide a systematic method for probing the details of atomic mechanisms responsible for aggregation. In this paper, the atomistic component of the overall framework is presented. First, a detailed set of targeted atomistic simulations are described that characterize fully the thermodynamic and transport properties of vacancy clusters over a wide range of sizes. It is shown that cluster diffusion is surprisingly favorable because of the availability of multiple, almost degenerate, configurations. A single large-scale parallel molecular dynamics simulation is then used to compute directly the evolution of the vacancy cluster size distribution in a supersaturated system initially containing 1000 uniformly distributed vacancies in a host lattice of 216,000 Si atoms at 1600 K. The results of this simulation are interpreted in the context of mean-field scaling theory based on the observed power-law evolution of the size distribution moments. It is shown that the molecular dynamics results for aggregation of vacancy clusters, particularly the evolution of the average cluster size, can be very well represented by a highly simplified mean-field model. A direct comparison to a detailed continuum model is made in a subsequent article. ABSTRACT A computational framework is presented for describing the nucleation and growth of vacancy clusters in crystalline silicon. The overall approach is based on a parametrically consistent comparison between two representations of the process in order to provide a systematic method for probing the details of atomic mechanisms responsible for aggregation. In this paper, the atomistic component of the overall framework is presented. First, a detailed set of targeted atomistic simulations are described that characterize fully the thermodynamic and transport properties of vacancy clusters over a wide range of sizes. It is shown that cluster diffusion is surprisingly favorable because of the availability of multiple, almost degenerate, configurations. A single large-scale parallel molecular dynamics simulation is then used to compute directly the evolution of the vacancy cluster size distribution in a supersaturated system initially containing 1000 uniformly distributed vacancies in a host lattice of 216,000 Si atoms at 1600 K. The results of this simulation are interpreted in the context of mean-field scaling theory based on the observed power-law evolution of the size distribution moments. It is shown that the molecular dynamics results for aggregation of vacancy clusters, particularly the evolution of the average cluster size, can be very well represented by a highly simplified mean-field model. A direct comparison to a detailed continuum model is made in a subsequent article.