Germanium dioxide (GeO2 ) takes two forms at ambient pressure: a thermodynamically stable rutile-type structure and a high-temperature quartz-type polymorph. Here, we investigate the phase stability at finite temperatures by ab initio phonon and thermochemical computations. We use gradient-corrected density-functional theory (PBE-GGA) and pay particular attention to the modeling of the "semicore" germanium 3d orbitals (ascribing them either to the core or to the valence region). The phase transition is predicted correctly in both cases, and computed heat capacities and entropies are in excellent agreement with thermochemical database values. Nonetheless, the computed formation energies of α-quartz-type GeO2 (and, consequently, the predicted transition temperatures) differ significantly depending on theoretical method. Remarkably, the simpler and cheaper computational approach produces seemingly better results, not worse. In our opinion, GeO2 is a nice test case that illustrates both possibilities and limitations of modern ab initio thermochemistry.