Latency

Where the seconds go between your camera and their screen

Chat replies to something you did eight seconds ago. That gap is built from six separate delays, and you own about two of them.

4 August 2026Latency

You wave. Chat waves back eight seconds later. People blame the internet.

The internet is a small part of it.

The stack

StageRough costYours?
Camera sensor to output1 to 2 framesNo
Capture into the encoder1 to 3 framesSome
Encoding, including lookahead and B-frames2 to 10 framesYes
Network and protocol buffer0.1 to 3 sYes
Ingest, transcode and packaging2 to 6 sNo
Player buffer2 to 10 sNo

Add it up. The two stages you cannot touch are the two biggest.

Why packaging costs so much

Platforms deliver over HTTP in segments. A player cannot start a segment that is still being written. So it waits for the whole thing, then holds a couple more in reserve against a stall.

Six-second segments with three buffered is eighteen seconds before anything else happens. Low-latency modes cut the segment into parts and ship each part as it finishes. That is where the drop from fifteen seconds to four comes from. It is not a faster network.

The one that surprises people

Bonding adds latency on purpose. An SRT buffer of 2000 ms is two seconds you agreed to spend so a dead modem does not end the stream.

What you can actually cut

The part worth accepting

Low latency and resilience pull against each other. Every buffer you remove is a buffer that was absorbing a problem.

Chat delay is annoying. A stream that drops out mid-sentence is worse. Pick the failure you would rather have, then set the buffers to match.

Sources

The Rig Wire reports on streaming technology from public documentation, source repositories and vendor releases. We did not bench-test what is described here. When we have tested something, we say so.