While GPU acceleration usually speeds up a simulation, it’s actually slower when running small models. This is related to the GPU architecture, which is less efficient whenever only a small fraction of the available threads can be utilized.