Forest dynamic models predict the current and future states of ecosystems and are a nexus between physiological processes and empirical data, forest plot inventories and remote-sensing information. The problem of biodiversity representation in these models has long been an impediment to a detailed understanding of ecosystem processes. This challenge is amplified in species-rich and high-carbon tropical forests. Here we describe an individual-based and spatially explicit forest growth simulator, TROLL, that integrates recent advances in plant physiology. Processes (carbon assimilation, allocation, reproduction, and mortality) are linked to species-specific functional traits, and the model was parameterized for an Amazonian tropical rainforest. We simulated a forest regeneration experiment from bare soil, and we validated it against observations at our sites. Simulated forest regeneration compared well with observations for stem densities, gross primary productivity, aboveground biomass, and floristic composition. After 500 years of regrowth, the simulated forest displayed structural characteristics similar to observations (e.g., leaf area index and trunk diameter distribution). We then assessed the model's sensitivity to a number of key model parameters: light extinction coefficient and carbon quantum yield, and to a lesser extent mortality rate, and carbon allocation, all influenced ecosystem features. To illustrate the potential of the approach, we tested whether variation in species richness and composition influenced ecosystem properties. Overall, species richness had a positive effect on ecosystem processes, but this effect was controlled by the identity of species rather by richness per se. Also, functional trait community means had a stronger effect than functional diversity on ecosystem processes. TROLL should be applicable to many tropical forests sites, and data requirement is tailored to ongoing trait collection efforts. Such a model should foster the dialogue between ecology and the vegetation modeling community, help improve the predictive power of models, and eventually better inform policy choices.