The development of effective strategies to improve the energy performance of the built environment depends on reliable data on the spatial and temporal distribution of energy demand and supply. As such, the interest in the urban energy computing has been steadily increasing. However, in most efforts, the informational and computational challenges have led to the adoption of simplified computational routines. These models fail to capture the temporal dynamics of load patterns and their dependency on transient phenomena (occupants and climate) with appropriate resolution. The present contribution reports on developmental activities towards generation of a bottom up urban stock heating demand model, which enables the use of Building Performance Simulation (BPS) tools for urban-level inquiries. For this purpose, a two-step method was adopted and applied to an urban instance in the city of Vienna, Austria. The first step, addresses the challenge of high informational and computational demand of BPS tools based on a systematic reduction of the extent of the required computations through sampling. Toward this end, key energy-relevant features of the buildings are used, along with a well-known datamining technique to classify the urban building stock and select representative buildings. Detailed descriptions of the selected buildings are utilized to generate detailed simulation models. Since loss of diversity is a natural consequence of any sample-based study, to recover part of the lost diversity, in a second step, a re-diversification routine was developed. This routine automatically generates permutations of the simulation models of the sample buildings, with diversified descriptions of non-geometric physical and operational building parameters. To represent operative diversity, stochastic techniques have been employed to model plausible yet diverse representations of occupants' presence and actions. The physical diversity, mainly pertaining to the thermal quality of construction components, has been treated through parametric representation of relevant material properties. As a prerequisite to the suggested method, GIS data and relevant performance assessment standards are utilized to generate an energy-relevant representation of the urban stock, which informs the two-step method. Since this framework reduces the computation domain in a first step and enhances it through the re-diversification process, the term “hourglass model” has been adopted to characterize it. The suggested method drastically reduces the modeling effort associated with large-scale application of BPS tools through sampling. Preliminary evaluations suggest a promising accord between the predicted and the expected values of heating demand, both at aggregated and disaggregated levels.