When using ffmpeg to record an .mp4 movie of a matplotlib animation with something like

anm.save('movie.mp4', writer='ffmpeg', extra_args=['-threads', 'n'])

where anm and n are the animation object and the number of threads, respectively, it seems that the global interpreter lock prevents from correctly distributing the task of encoding among available cpu cores.

Of course, one can avoid this by first breaking the animation into n chunks and encoding each using ffmpeg with, for example, Python's multiprocessing library, and then stitching the resulting 'n' movies into a single one.

Would the above be the best solution, or should perhaps matplotlib try to handle this internally?