Detailed and convincing walkthrough auralizations of virtual rooms requires much processing capability. One method of reducing this requirement is to pre-calculate a data-set of room impulse responses (RIR) at locations throughout the space. Processing resources may then focus on RIR interpolation and convolution using the dataset as the virtual listening position changes in real-time. Recent work identified the suitability of wave-based models over traditional ray-based approaches for walkthrough auralization. Despite the computational saving of wave-based methods to generate the RIR dataset, processing times are still long. This paper presents a wave-based implementation for execution on a general purpose graphics processing unit. Results validate the approach and show that parallelisation provides a notable acceleration.