A reference frame can be used by a P-frame, also known as a predictive or partial frame, to help define a future frame in a compressed video. While H.264 can support any number of reference frames, we recommend using between three and five (three to five reference frames can be used by a P-frame) for optimal performance across devices.
H.264 levels limit the decoded picture buffer. For example, if level 4.1 is being used, then the maximum reference frames allowed for 720p and 1080p videos are nine and four, respectively. It is important to keep the number of reference frames below the allowed H.264 level limits of the targeted device. For more details, you can view the full H.264 profile and level limitations table here: