Mantle flow induces dynamic topography at the core-mantle boundary (CMB), with distribution and amplitude that depend on details of the flow. To assess whether observations of CMB topography can give constraints on deep mantle structure, we determine CMB dynamic topography associated with different models of mantle convection, including thermochemical and purely thermal models. We investigate the influence of key controlling parameters, specifically the thermal viscosity ratio (Delta eta(T)) and, for thermochemical models, the density contrast (Delta rho(C)) and viscosity ratio (Delta eta(C)) between primordial and regular materials. In purely thermal models, plume clusters induce positive topography with an amplitude that decreases with increasing Delta eta(T). In thermochemical models with moderate density contrasts, around 100-200 kg m(-3), reservoirs of dense material induce depressions in CMB topography, surrounded by a ridge of positive topography. The average depression depth and ridge height increase with increasing Delta rho(C) and Delta eta(C), but decrease with increasing Delta eta(T). We find that for purely thermal models or thermochemical models with Delta rho(C) similar to 90 kg m(-3) and less, the long-wavelength (spherical harmonic degrees up to l = 4) dynamic topography and shear wave velocity anomalies predicted by thermochemical distributions anticorrelate. By contrast, for models with Delta rho(C) >= 100 kg m(-3) and Delta eta(C) > 1, long-wavelength dynamic topography and shear wave velocity anomalies correlate well. This potentially provides a test to infer the nature, that is, either purely or mostly thermal (Delta rho(C) <= 100 kg m(-3) m(-3)) or strongly thermochemical (Delta rho(C) >= 100 kg m(-3)), of the low shear wave velocity provinces observed by global tomographic images. The presence of post-perovskite, provided that its viscosity is similar to that of bridgmanite, does not alter these conclusions.