Alright, @farhana_hossain. So, it is not that backprop is faster than forward prop.
Instead, it is that computing gradients with backprop ( N + P ) is faster than computing gradients without backprop ( N x P ). Please watch the video again for getting the idea that it is not a comparison between forward prop and backprop, but a comparison between computing gradients with and without backprop.
As for why backprop is faster, I suggest this post.
Cheers,
Raymond