What is the difference between TCP and UDP?
TCP (Transmission Control Protocol) numbers every byte and resends anything lost, so data arrives complete and in order. UDP (User Datagram Protocol) skips the checks and just sends, so any repair is left to the app.

TCP vs UDP
TCP (Transmission Control Protocol) is one of the two main rulebooks that carry data across the internet. It is the one that promises delivery: every byte you send is numbered, the sender keeps a copy until the other side confirms it arrived, and anything lost is sent again. That is why a photo sent from a train with terrible signal can still arrive without one broken pixel.
UDP (User Datagram Protocol) is the other rulebook, and it exists because not everything needs that promise. Some exchanges are a single question and a single answer, and a handshake plus confirmations would cost more than they save. UDP strips all of that away and sends the data with a tiny header in front. If anything needs checking, the app on top does it.
What TCP actually does
Before any data moves, both sides shake hands. Your phone sends a SYN, the server answers SYN-ACK, and your phone replies ACK. Only then does the photo start to flow. Each chunk travels with a TCP header of at least 20 bytes, and that header carries a sequence number. The number is not a chunk count: TCP numbers bytes, so the sequence number is the position of the chunk’s first byte in the photo. The receiver uses it to put the chunks in order and to spot a gap.
Each chunk that lands triggers an acknowledgment going back. The ACK names the next byte the server expects, so one acknowledgment confirms everything up to that point. If a chunk goes missing, the server keeps answering later chunks with the same ACK, still asking for that same byte. After three of those duplicate ACKs your phone resends the missing chunk from the copy it kept, without waiting. If even that signal never comes, a retransmission timer runs out and the chunk is resent anyway. Either way, the server ends up with every byte, in order.
How UDP works without the checks
- No handshake at all. A UDP packet just leaves. There is no
SYN, noSYN-ACK, and noACKbefore the first byte of data moves. - An 8-byte header only. The whole UDP header is four fields: source port, destination port, length, and a checksum. Then comes the data.
- No numbers, no receipts. UDP has no sequence numbers and no acknowledgments, so nothing is put back in order and nothing is resent by UDP itself.
- The app checks if it wants to. HTTP/3 runs on QUIC, which runs on UDP and adds its own packet numbering, acknowledgments, and loss recovery on top.
The result is speed. A DNS (Domain Name System) lookup uses UDP on port 53: one question, one answer, one trip. Recent versions of the dig tool even print (UDP) after the SERVER line to show it.
UDP is not the broken option
People hear “no confirmations” and assume UDP is the careless choice. It isn’t. UDP simply moves the checking out of the transport and into the app, where the app decides what it needs. HTTP/3 is the clearest proof: it delivers web pages, images, and video reliably, and it does so over UDP because QUIC handles the numbering and resending itself. Meta reports that more than 75 percent of its internet traffic uses QUIC and HTTP/3, including Instagram on iOS and Android. So the photo from the train may well have been repaired by QUIC rather than TCP. The idea is the same either way: number the data, confirm what arrived, resend what didn’t.
What to do
- Run
digfor a name you know and look for(UDP)after the SERVER line: that is a real UDP port 53 lookup in front of you. - When you look at an app, ask which rulebook it needs: every byte in order (TCP, or QUIC on top of UDP) or one fast trip (plain UDP).
- Remember the sizes: a TCP header is at least 20 bytes, a UDP header exactly 8. That difference is the price of the promise.
- Ask the video’s question about your own video calls: do they need every byte in order, or do they need speed? The answer tells you which rulebook fits.
Watch it in 60 seconds
