A 100 Mbps connection transfers about 12.5 megabytes a second. Nothing is being throttled: the line is sold in megabits and the file is measured in megabytes, and there are eight bits in a byte.
The lowercase b and the uppercase B are the entire difference, and they are one keystroke apart in every specification anyone reads.
Why are the two units different at all?
History, and it has stuck. Transmission has always been counted in bits because that is what a line carries one at a time, while storage has been counted in bytes because that is the smallest addressable unit.
| Line rate | Bytes per second | 1 GB file |
|---|---|---|
| 10 Mbps | 1.25 MB/s | ~13 min |
| 100 Mbps | 12.5 MB/s | ~80 s |
| 500 Mbps | 62.5 MB/s | ~16 s |
| 1 Gbps | 125 MB/s | ~8 s |
Those times are the theoretical best and nothing achieves them. Dividing the advertised number by eight is the first correction; the rest of this article is the other corrections.
The second unit trap sits underneath. A drive maker counts a gigabyte as a thousand million bytes and an operating system has traditionally counted it as 1,073,741,824, so the same file has two sizes depending on who is reporting it.
Where does the rest of the capacity go?
Into headers. Every packet carries addressing and control information that is not your data, and the framing underneath carries more — so the useful throughput on a link is meaningfully below its line rate before anything else goes wrong.
Retransmission takes another slice. A lost packet is sent again, and the protocol slows down when it detects loss, so a link with a small error rate loses more throughput than the error rate alone suggests.
The practical figure is that a healthy connection delivers somewhat less than its advertised rate in real transfers, and a transfer that reaches most of the theoretical maximum is doing well rather than badly.
Why does more bandwidth not feel faster?
Because the thing that made it feel slow was latency, and the two are unrelated. Bandwidth is how much fits through per second; latency is how long one round trip takes.
Loading a page involves many small round trips — a name lookup, a connection, a request, a response — and each one costs a full journey there and back regardless of capacity. Doubling the capacity does not shorten the journey.
There is a floor set by physics. Light travels through fibre at roughly two thirds of its speed in vacuum, around 200,000 kilometres a second, so a 5,600-kilometre link has a one-way delay of about 28 milliseconds and a round trip of about 55 before any equipment is involved. Real routes are longer than the straight line, so 70 to 80 milliseconds across the Atlantic is close to as good as it gets.
What is the bandwidth-delay product?
The amount of data in flight on a link at any moment: capacity multiplied by round-trip time. It is the number that explains why a fast, distant link can transfer slowly.
A protocol only sends so much before waiting for acknowledgement. If that allowance is smaller than the bandwidth-delay product, the sender spends most of its time idle — the pipe is wide, and the sender keeps stopping to check the far end is still there.
That is why a single transfer over a long distance often underperforms while several parallel transfers saturate the same link. Each stream has its own allowance, and together they fill the pipe.
Congestion is the third factor and the one that varies by hour. A shared segment carries everyone’s traffic, so the evening figure and the mid-morning figure on the same line can differ substantially without anything being faulty. Measuring twice at different times says more than measuring once.
Why is upload slower than download?
Because most residential connections are deliberately asymmetric. Capacity is split in favour of the direction people use most, so a line advertised at 100 Mbps down may offer a fraction of that up.
It stops being a detail the moment you send rather than receive: backups, video calls, and uploading footage all run at the smaller number. A cloud backup estimate based on the download figure will be wrong by the ratio between them.
Symmetric connections exist and are usually a business product, which is the honest reason they cost more for the same headline number.
Questions people ask
Do I need gigabit at home? Only if several people transfer large files at once. For streaming and browsing the limit is almost always latency and Wi-Fi.
Is a speed test the real number? It measures the path to that server at that moment. Two tests to different servers disagreeing is normal.
Does Wi-Fi change this? Yes, and usually it is the limit. A wireless link is shared and variable in a way the line behind it is not.
What about MB/s versus MiB/s? The second is the binary megabyte, about five per cent larger. Transfer tools disagree about which they report.
Why is my transfer slower than both ends? The bottleneck is whichever link in the chain is narrowest, and it is often neither end.
Do the division before you plan around a number. The bandwidth calculator sizes a connection against a workload, download time and upload time turn a file size into a duration, the latency calculator covers the part more capacity cannot fix, and Mbps to MB/s and MB/s to Mbps settle the factor of eight in either direction.