Data Units: KB vs KiB, and Why Disks Look Smaller

A bit is a single 0 or 1; eight bits make a byte, the basic unit of storage. The trouble starts with the prefixes. Computers count in binary, where powers of two are natural, so 2^10 = 1024 became "one kilobyte" in early computing. But the metric kilo- officially means exactly 1000. So the same abbreviation, KB, came to mean 1024 bytes in one context and 1000 bytes in another.

To end the ambiguity, the IEC introduced binary prefixes in 1998: kibibyte (KiB) = 1024 bytes, mebibyte (MiB) = 1024 KiB, gibibyte (GiB) = 1024 MiB, and so on. Under this scheme, decimal KB/MB/GB mean powers of 1000, while binary KiB/MiB/GiB mean powers of 1024. The gap widens as units grow: it is about 2.4% at the kilobyte level but roughly 10% by the terabyte level — which is exactly why a 1 TB (decimal) drive shows as about 0.909 TiB, or 931 GiB.

In practice, storage manufacturers advertise in decimal units because the numbers look bigger, while Windows and many tools report in binary units but label them GB. macOS and most drive labels use decimal. Networking, meanwhile, measures speed in decimal bits per second — a 100 Mbps link is 100,000,000 bits/s.

If you are estimating download times or reading a marketing spec, decimal is usually intended. If you are checking file sizes in an operating system, binary is likely. UnitBook's data category uses binary steps (1 KB = 1024 bytes), matching how most operating systems display sizes, so its numbers line up with what you see in your file manager.

A drive sold as 1 TB really does hold 1 000 000 000 000 bytes — the manufacturer is using SI prefixes correctly. Windows then divides that by 2³⁰ and shows 931 GB, because it is counting in gibibytes while printing the letters GB. Nothing has gone missing; the same quantity has simply been measured with a different-sized ruler. The same arithmetic turns a 500 GB drive into 465 GB and a 2 TB drive into 1.81 TB.

On top of the unit mismatch sit real losses. Formatting writes filesystem metadata; ext4 reserves five per cent of the volume for root by default; a laptop usually ships with a recovery partition and a preinstalled operating system; and every file rounds up to a whole allocation cluster, so a folder of thousands of tiny files occupies noticeably more than the sum of its file sizes.

Solid-state drives take one more slice. Controllers hold back a portion of the flash — commonly seven to twenty-eight per cent — as over-provisioning for wear levelling and garbage collection. That space never appears to the operating system at all, which is why some enterprise SSDs are sold as 960 GB rather than the 1 TB their chips add up to.

Knowing the habits of each platform saves a lot of confusion. Windows reports binary quantities under decimal labels. macOS switched to honest decimal reporting in Snow Leopard in 2009, and iOS followed, which is why the same drive reads differently on two machines. On Linux, `df -h` and `ls -h` are binary while `df -H` and `ls --si` are decimal, and most distributions print the unambiguous KiB/MiB/GiB spellings.

Memory is the one place where binary is always right. A 16 GB memory module really is 16 GiB, because addressing works in powers of two and a chip cannot be built with a round decimal capacity. CPU caches, page sizes and sector sizes follow the same rule, which is why you see 4 KiB pages and 64-byte cache lines rather than tidy decimal numbers.

Networking never went binary at all. A 300 Mbps link means 300 000 000 bits per second, decimal, and bits rather than bytes — so the ceiling on a download is about 37.5 MB/s before protocol overhead. Mixing the two conventions is what makes a fast connection feel eight times slower than the number on the contract.