The Wide Field InfraRed Survey Telescope (WFIRST) is the next NASA astrophysics flagship mission, to follow the James Webb Space Telescope (JWST). The WFIRST mission was chosen as the top-priority large space mission of the 2010 astronomy and astrophysics decadal survey in order to achieve three primary goals: to study dark energy via a wide-field imaging survey, to study exoplanets via a microlensing survey, and to enable a guest observer program. Here we assess the ability of the several WFIRST designs to achieve the goal of the microlensing survey to discover a large sample of cold, low-mass exoplanets with semimajor axes beyond roughly one AU, which are largely impossible to detect with any other technique. We present the results of a suite of simulations that span the full range of the proposed WFIRST architectures, from the original design envisioned by the decadal survey, to the current design, which utilizes a 2.4-m telescope donated to NASA. By studying such a broad range of architectures, we are able to determine the impact of design trades on the expected yields of detected exoplanets. In estimating the yields we take particular care to ensure that our assumed Galactic model predicts microlensing event rates that match observations, consider the impact that inaccuracies in the Galactic model might have on the yields, and ensure that numerical errors in lightcurve computations do not bias the yields for the smallest mass exoplanets. For the nominal baseline WFIRST design and a fiducial planet mass function, we predict that a total of ${∼}1400$ bound exoplanets with mass greater than ${∼}0.1~M_{⊕}$ should be detected, including ${∼}200$ with mass ${\lesssim}3~M_{⊕}$. WFIRST should have sensitivity to planets with mass down to ${∼}0.02~M_{⊕}$, or roughly the mass of Ganymede.