Smart cities are providing advanced services gathering data from different sources. Cities collect static data like road graphs, service description as well as dynamic/real time data like weather forecast, traffic sensors, bus positions, events, emergency data, etc. RDF stores may be used to integrate all information coming from different sources and allow applications to use the data to provide new advanced services to the citizens and city administrators exploiting inferential capabilities. These city services are typically based on geographic positions and need to access quickly to the real time data (e.g., next time of bus arrival) as well as to the historical data to perform some data analysis to compute predictions. In this paper, the needs and constraints for RDF stores to be used for smart cities services and the currently available RDF stores are evaluated. The assessment model allows understanding if they are suitable as a basis for Smart City modeling and application. The benchmark proposed has been defined for generic smart city services to compare results that can be obtained using different RDF Stores. In the benchmark, particular emphasis is devoted to geo and full text searches that are partially considered in other well-known RDF store benchmarks as LUBM and BSBM. The paper reports the validation of the proposed Smart City RDF Benchmark (http://www.disit.org/smartcityrdfbenchmark) on the basis of Florence Smart City accessible as Km4City. The comparison addressed a number of well-known RDF stores as Virtuoso, GraphDB and many others.