This paper explores the application of Gaussian process tomography (GPT) to the bolometer diagnostic at ASDEX Upgrade. Previous work is extended by nonstationary Cartesian kernels and kernels that incorporate magnetic equilibrium information. With the help of existing tomograms from actual shots, a set of hyperparameters working well on a broad range of bolometer measurements is found. While other tomography methods usually rely on additional information like the magnetic equilibrium, those are not required by the GPT method, and it is shown that the method presented here can produce good results without additional information. In addition, the real-time capability of this approach is successfully demonstrated by computing tomograms for 10s of bolometer data in under 5s. This is possible without providing additional information besides bolometer measurements and enables the possibility of real-time reactor control for the future.