Abstract Increasing population and urbanisation threaten sustainable urban development due to increased resource consumption and emissions. As buildings are one of the largest energy consumers, it is crucial that their thermal load can be inspected on a large scale and at the highest resolution possible. The proposed method is performed in two stages. First, the LiDAR data and buildings’ metadata are preprocessed to generate high-resolution 3D building models that are represented by a triangle mesh. Thermal load of buildings throughout the year is then calculated per-triangle in a parallelised manner, while considering local micro-climate and shadowing from surroundings. Parallel design of the estimation enables significant speed-up of large-scale workloads, while maintaining accurate shadowing estimation. In experiments, the method was applied over a part of the city of Maribor, where heating and cooling loads were inspected in addition to other factors of thermal load estimation. Yearly thermal load calculation with an hourly time-step for 4,817 buildings with over 9.17 million triangles took about 8 min on a modern GPU. When comparing the run-times using a GPU and a modern CPU, the GPU was more than 60-times faster than a CPU for a million triangles. The speed-up grew with the number of triangles.