An ISP quotes bits and your computer shows bytes. That is a factor of eight — and then it happens again, more quietly.
Calculator
What internet connections are sold in, worldwide. The number on the bill.
Try:
100 Mbps is
12.50 MB/s
…and 11.92 MiB/s, which is what a file manager will show. Those two are different numbers for the same speed.
Two errors, not one
Bits against bytes
÷ 8
Decimal against binary
− 4.86%
Combined
÷ 8.3886
Reading the connection’s number as megabytes is out by eight. Having fixed that, a file manager showing MiB/s is out by a further 4.86%, because a mebibyte is 1,048,576 bytes rather than a million. The two compound to exactly 8.388608 — which is 8 × 1.048576, and why the number on the bill and the number on the screen never look related.
Every unit
Bits per second100,000,000bit/s · bits
Kilobits per second100,000kbps · bits
Megabits per second100.00Mbps · bits
Gigabits per second0.1000Gbps · bits
Terabits per second1.00e-4Tbps · bits
Kilobytes per second12,500kB/s · bytes
Megabytes per second12.50MB/s · bytes
Gigabytes per second0.0125GB/s · bytes
Kibibytes per second12,207KiB/s · bytes, binary
Mebibytes per second11.92MiB/s · bytes, binary
Gibibytes per second0.0116GiB/s · bytes, binary
Each row says whether it counts bits or bytes, and whether it is decimal or binary — because those two facts, not the prefix, are what make two of these numbers differ.
How long a file takes
Headers plus acknowledgements, retransmission and the gap between line rate and sustained throughput.
Theoretical6 min 40 s400 s
With 8% overhead7 min 15 s435 s
Every other calculator gives the theoretical figure, and it is never achievable: protocol headers consume a few percent before any payload moves, and that is before congestion, retransmission or a slow server at the far end. Treat the theoretical time as a floor rather than an estimate.
What doubling your speed would buy
Now6 min 40 s
At twice the speed3 min 20 s
Saved3.3 min50% either way
Doubling always halves the time — that percentage never changes. What changes is how much time that is worth. On this file, doubling from 10 Mbps would save 33 minutes and doubling from 100 would save three. Time is reciprocal in speed, so upgrades buy progressively less the faster you already are.
What this converter covers
Shows the two errors separately, because knowing there are two is what makes the answer stick.
Mbps, Gbps, MB/s, MiB/s and the rest, each labelled bits or bytes
The eightfold error and the 4.86% one, stated apart
Download times, theoretical and with protocol overhead
What doubling your connection would actually save
Familiar line rates for comparison
Two separate errors Bits and bytes Decimal and binary Real download times
100 Mbps = 12.5 MB/s = 11.92 MiB/s · a 5 GB file takes 6 min 40 s at best
Educational estimate
Planning support from the values you enter — not professional advice.
Megabits are not megabytes
Internet connections are sold in bits per second. Files, drives and download progress bars are measured in bytes. A byte is eight bits, so the two numbers differ by a factor of eight — and the only thing distinguishing them in writing is the case of a single letter.
What each advertised connection can actually move
Sold as
Maximum MB/s
A 5 GB file
8 Mbps
1.0
1 h 23 min
20 Mbps
2.5
33 min
100 Mbps
12.5
6 min 40 s
500 Mbps
62.5
1 min 20 s
1000 Mbps
125
40 s
This is why a hundred-megabit connection shows a download running at twelve and a half, and why that is not a fault. It is the connection working exactly as sold, described in a different unit.
The convention is not arbitrary. Networking has always counted bits because that is what actually travels on the wire, and a link’s capacity is a property of the signalling rather than of the data. Storage counts bytes because that is the smallest addressable unit. Both are reasonable, and the collision happens at exactly the point a customer looks at both numbers.
The one convenient case worth remembering: 8 Mbps is exactly 1 MB/s. Anything else can be divided by eight from there.
And then it happens again
Having divided by eight, you have 12.5 MB/s. Open a file manager during that download and it will very likely say 11.9. That is a second, separate error, and almost nobody mentions it.
A megabyte in the SI sense is a million bytes. A mebibyte — MiB — is 1,048,576, because it is 2²⁰. Software has historically counted in powers of two while marketing counts in powers of ten, so the same speed reads 4.86% lower when shown in MiB/s.
One connection, three correct numbers
Sold as
MB/s (decimal)
MiB/s (binary)
Gap
100 Mbps
12.50
11.92
4.86%
500 Mbps
62.50
59.60
4.86%
1 Gbps
125.00
119.21
4.86%
So there are two errors between the bill and the screen, not one. Together they come to exactly 8 × 1.048576 = 8.388608, and that is why “100” and “11.9” look completely unrelated: the ratio between them is not a round number in either system.
The calculator reports both factors separately for that reason. Knowing the answer is 11.9 solves one download; knowing there are two distinct conversions between the two numbers solves the question permanently.
Line rate is not throughput
Even at 12.5 MB/s exactly, a 5 GB file will not arrive in six minutes and forty seconds. Some of the capacity carries the packaging rather than the contents.
Every packet carries headers. Over Ethernet with TCP and IPv4 that is 20 bytes of IP header, 20 of TCP and the Ethernet framing around it — about 3.2% of a full-size frame. IPv6 headers are 40 bytes rather than 20, so the same traffic costs about 4.5%.
A 5 GB file on a 100 Mbps connection
Assumption
Effective rate
Time
Theoretical
12.50 MB/s
6 min 40 s
TCP/IPv4 headers
12.11 MB/s
6 min 53 s
TCP/IPv6 headers
11.94 MB/s
6 min 59 s
Typical practical
11.50 MB/s
7 min 15 s
The header cost is the part that can be calculated. Everything else that slows a real transfer — congestion, a busy server, packet loss forcing TCP to back off, Wi-Fi, the write speed of the destination drive — can only be measured, which is why this page offers a floor rather than an estimate.
A small coincidence worth noticing: the IPv6 overhead of about 4.5% is close to the 4.86% between MB/s and MiB/s. So a practical throughput in MiB/s often lands near the naive theoretical figure, for two entirely unrelated reasons that happen to cancel. It is a coincidence rather than a rule, and it is worth knowing precisely so that it does not get mistaken for one.
Why upgrades buy less as you go faster
Time is inversely proportional to speed, which means doubling always halves the time — and that the amount of time saved collapses as the connection gets faster.
Doubling the connection, on a 5 GB file
Upgrade
Before
After
Saved
10 → 20 Mbps
67 min
33 min
33 min
50 → 100 Mbps
13 min
7 min
7 min
100 → 200 Mbps
7 min
3 min
3 min
500 → 1000 Mbps
80 s
40 s
40 s
Every row halves the time — the percentage is identical. But the first row saves half an hour and the last saves forty seconds. The same upgrade, fifty times less benefit.
That is the practical case against buying speed you do not need. Going from 10 to 100 Mbps is transformative; going from 500 to 1000 changes a forty-second download into a twenty-second one, and at that point the bottleneck has usually moved elsewhere anyway — to the server, the Wi-Fi, or the drive.
This is the same reciprocal shape as fuel economy, where 10 to 20 mpg saves far more fuel than 30 to 50, and as tempo, where ten beats per minute is worth seven times more at the slow end. Whenever a quantity is expressed per something, equal steps in the rate are curved steps in what you actually care about.
Reading a speed test honestly
A few things make speed figures easier to interpret, most of them about knowing which number is being reported.
Speed tests report bits; downloads report bytes. A test showing 94 Mbps and a browser showing 11 MB/s are the same measurement. Divide by eight before comparing anything.
A single download rarely uses the whole connection. One TCP stream is limited by latency as well as bandwidth, which is why a download manager opening several connections is often genuinely faster on the same line.
Check the destination drive. A gigabit connection delivers 125 MB/s, which a mechanical hard drive cannot sustain while also doing anything else. Above a few hundred megabits the drive becomes the limit before the line does.
Upload is usually the smaller number, often by a large factor on cable and DSL. Video calls, backups and sending files are limited by it, and it is the figure most people never check.
One last habit worth having: when comparing two quoted speeds, convert both into the time for a file you actually move. “300 against 500 Mbps” is abstract; “two minutes against seventy seconds for the weekly backup” is a decision.
Related calculators
Related units, and the other pages where a rate is reciprocal:
Data StorageBytes, KB, MB, GB and TB alongside the binary KiB, MiB, GiB and TiB — and why a 1 TB drive shows as 931 GB.
Numeral BaseBinary, octal, decimal and hex — showing the 4-bit grouping that makes hex work, and handling negatives at a width you choose.
Fuel Economympg, L/100 km and km/L — US and UK mpg kept apart, and the reciprocal explained, since a better car has a LOWER L/100 km.
Tempo and Delay TimeBPM to delay and reverb times in ms and Hz, with straight, dotted and triplet side by side — and the feedback tail.
Scientific NotationScientific, engineering and decimal forms with significant figures counted — and ambiguous inputs flagged rather than silently resolved.
Frequency to WavelengthFrequency, wavelength, photon energy and wavenumber — in vacuum and in the medium, including coax velocity factors.
The binary prefixes are standardised, which is what makes MiB/s and MB/s different units rather than different house styles — the 4.86% is a consequence of the standard rather than an approximation. The protocol overheads come from the header sizes the relevant RFCs specify, and are a floor on the cost rather than a prediction about any particular network.
IEC 80000-13 — Information science and technology: binary prefixesInternational Electrotechnical Commission · verified 2026-09-07 · The kibi, mebi and gibi prefixes as powers of 1024, distinct from the SI kilo, mega and giga as powers of 1000 — the standard that makes MiB/s and MB/s genuinely different units rather than a matter of style
NIST — prefixes for binary multiplesNational Institute of Standards and Technology, US · verified 2026-09-07 · That the SI prefixes mean powers of ten and should not be used for powers of two, which is why the 4.86% gap between MB/s and MiB/s exists at all
RFC 791 and RFC 8200 — Internet Protocol, versions 4 and 6Internet Engineering Task Force · verified 2026-09-07 · The IPv4 and IPv6 header sizes of 20 and 40 bytes behind the protocol overhead figures — and the reason IPv6 costs about a percentage point and a half more than IPv4
Conversion note
These are arithmetic conversions and a best case, not a prediction of what a transfer will do. Real throughput depends on far more than the line rate: congestion along the path, the capacity and load of the server at the other end, latency and packet loss interacting with TCP's window, Wi-Fi conditions, the speed of the drive you are writing to, and any traffic shaping or data-cap throttling your provider applies. Advertised connection speeds are typically maximums rather than guarantees, and the figure that matters for most real use is what the connection sustains at a busy time of day rather than its peak. The overhead figures here account for protocol headers only, which is the part that can be calculated; everything else in that list can only be measured. Use the theoretical time as a floor, the overhead time as a better floor, and a real transfer as the only actual answer.
Published the Data Transfer Rate Converter: Mbps, Gbps, MB/s, MiB/s and the rest, each labelled bits or bytes and decimal or binary.
Separates the two errors between an ISP's number and a file manager's rather than giving only the answer. Reading Mbps as MB/s is a factor of eight; a file manager showing MiB/s is a further 4.86 percent, because a mebibyte is 1,048,576 bytes rather than a million.
Notes that the two compound to exactly 8.388608, which is 8 times 1.048576 -- and why 100 Mbps and 11.9 MiB/s look completely unrelated.
Gives download times both theoretically and with protocol overhead, since every other calculator stops at the theoretical figure and it is never achievable: TCP/IPv4 headers cost about 3.2 percent and IPv6 about 4.5.
Explains that transfer time is reciprocal in rate, so doubling always halves the time while the amount of time saved collapses -- 33 minutes going from 10 to 20 Mbps against 3 minutes going from 100 to 200.
Cross-links that reciprocal shape to fuel economy and tempo, where the same arithmetic produces the same misunderstanding.
Flags the coincidence that IPv6 overhead of about 4.5 percent nearly cancels the 4.86 percent MB-to-MiB gap, and says explicitly that it is a coincidence rather than a rule.
Deliberately scoped away from the Data Storage Converter, which owns capacities: this page converts rates.
Verified by 66 automated cases, including that every unit's bits-or-bytes and decimal-or-binary flags match its actual arithmetic, and that overhead lengthens a transfer by exactly 1/(1 minus p).
Add this calculator to your site
Responsive embed — and private: nothing your visitors type leaves their browser.