EMZETT.Bandwidth
In short: The maximum amount of data a connection can transfer per unit of time, stated e.g. in Mbit/s or Gbit/s.
In more detail: Bandwidth describes the theoretical capacity of a line — how much actually arrives also depends on latency, data loss and utilisation. It’s often equated with speed, but strictly speaking it’s only one factor of it (comparable to the number of lanes on a motorway, not the speed you drive at).
In Depth
Bit vs. byte — the most common confusion
Bandwidth is measured in bits per second (Mbit/s, Gbit/s) — the important distinction is from bytes per second (MB/s, GB/s), which is often shown for download speeds in download managers or browsers: 1 byte = 8 bits, so an internet connection with “100 Mbit/s” ideally downloads at a maximum of 12.5 MB/s, not 100 MB/s. This confusion is one of the most common reasons for “my line is slower than advertised” bewilderment among customers who booked their plan in Mbit/s but read the download speed in MB/s and forget to divide by eight.
Bandwidth, throughput and goodput
Network engineers distinguish several related but different terms: bandwidth is the theoretical capacity of a line, throughput the data rate actually achieved including protocol overhead, and “goodput” the rate of actually usable application data after deducting headers, retransmissions and encryption overhead. On real connections these three almost always diverge in this order — bandwidth > throughput > goodput. The real, usable bandwidth is almost always well below the theoretical maximum for several reasons: protocol overhead (every packet carries header information for TCP/IP, Ethernet frames etc. alongside the payload), shared use of the line by several devices, as well as losses from data loss and the retransmissions this makes necessary.
Symmetric vs. asymmetric
Many classic internet connections (DSL, cable) are asymmetric: the download bandwidth is considerably higher than the upload bandwidth, because most private users have historically downloaded much more (web pages, streaming) than they upload. Fibre connections (FTTH), on the other hand, often offer symmetric bandwidth — relevant for anything that serves data itself, such as video calls, cloud backups or your own server.
Shared media: Wi-Fi vs. Ethernet
With Wi-Fi there’s the added factor that all devices within radio range share the same bandwidth — the more simultaneously active devices, the less is left per device, quite unlike a dedicated Ethernet connection, where every cable (or every switch port) is its own full-duplex segment. A café with a single Wi-Fi router and 50 simultaneously active users feels this much more strongly than an office with 50 individually wired Ethernet ports on the same switch.
Bandwidth vs. latency — not the same thing
In practice it’s also crucial to separate bandwidth from latency: a line can have a very high bandwidth (many lanes on the motorway) and still have high latency (the individual journey takes a long time, e.g. with satellite internet over long distances, where the signal literally has to travel to the satellite and back). For video streaming, sufficient bandwidth is what counts most (the amount of data has to arrive in time; small delays are compensated for by buffering), whereas for online gaming or video calls, low latency is what matters most (every delay is directly noticeable as stuttering or dropouts, no matter how much bandwidth is available).
Measuring in practice
Tools like Speedtest primarily measure bandwidth via a short, intensive test transfer, while tools like iperf are used specifically for deeper network diagnostics, for example to check the actual capacity of an internal connection between two servers, independent of the internet connection.