This improves detection by performing an actual check on encoder capability directly with the NVENC/CUDA libraries.