Standard GPUs absolutely do. Since CUDA 11, all CUDA cards expose the same featureset at differing speeds (based on backing capability). You can absolutely (try to) run CUDA UMA on your 2060, and it will complete the computation.