This is very inefficient due to Amdahl's law. "-parallel" causes the encoder to instead spawn a single job per input texture, which can be far more efficient. -opencl and -parallel can be used ...