Byte
Non-system unit · quantity: amount of information
01 · Definition
The word was coined by the IBM engineer Werner Buchholz in 1956, deliberately distorting bite so that it would not be confused with bit. Eight bits to a byte took hold only with System/360 in 1964.
A byte is eight bits, that is 256 distinguishable states. It is the smallest addressable piece of memory in any modern processor: a single bit cannot be read, the whole cell is read. Eight was chosen not by physics but by convenience: a power of two, enough for the Latin alphabet with its digits and signs, divisible into two hexadecimal digits. A byte holds an ASCII character, but not a Cyrillic letter in UTF-8 — there two are needed, and an emoji takes four.
In UTF-8 a Latin letter takes one byte, a Cyrillic one two, an ideograph three, an emoji four. A character is therefore not a byte: the length of a string in characters and its length in bytes are different numbers, and confusing them is the commonest source of broken text.
02 · Conversion
Enter a volume — the sheet will convert it
The byte inherited its double system of prefixes from the bit. The decimal kilobyte is a thousand bytes, the binary kibibyte 1024. Disk makers count in decimal ones, operating systems in binary — hence the «missing» gigabytes on every new drive.
Disks worked for decades in sectors of 512 bytes, modern ones in 4096. A file system does not read less than a sector, so a file of one byte still occupies a whole one.
| Unit | Name | Value | Where it is met |
|---|---|---|---|
| B | byte, eight bits | 1000 | the main notation |
| bit | bit | 8000 | channel, speed |
| kB | kilobyte, 1000 bytes | 1 | decimal prefix |
| MB | megabyte, 10⁶ bytes | 0.001 | files, photographs |
| GB | gigabyte, 10⁹ bytes | 1·10⁻⁶ | disks, media |
| TB | terabyte, 10¹² bytes | 1·10⁻⁹ | storage |
| KiB | kibibyte, 1024 bytes | 0.9766 | memory, file systems |
| MiB | mebibyte, 1024 kibibytes | 9.537·10⁻⁴ | working memory |
| GiB | gibibyte, 1024 mebibytes | 9.313·10⁻⁷ | operating systems |
| TiB | tebibyte, 1024 gibibytes | 9.095·10⁻¹⁰ | disk arrays |
| B6 | six-bit byte | 1333 | before System/360 |
| B | byte | 1000 | the modern notation |
Decimal prefixes take a thousand, binary ones take 1024 and are written with an «i»: KiB, MiB, GiB. A disk of one terabyte holds 10¹² bytes, and the system shows them as 931 GiB: both sides count correctly, simply in different prefixes.
03 · Orders of magnitude
from a character to a data centreA photograph — 4 MB
04 · Measuring instruments
What bytes are counted with
The capacity on the box is counted in decimal gigabytes, the file system counts in binary ones. Neither side is deceiving anyone, but the buyer sees a smaller number than the one printed on the label.
Here the binary prefixes are honest: a module of 16 GB contains exactly 16 gibibytes. Memory is addressed in powers of two, and no other counting is possible for it.
The provider sells megabits, the downloader shows megabytes. Divide by eight: 100 Mbit/s give about 12 MB/s.
It shows a file byte by byte: codes on the left, characters on the right. That is how one reads formats without documentation and finds out where the header ends.
05 · Writing rules
B capital, the k in kilobyte lower-case
The decimal kilobyte is written kB with a lower-case k, like every SI prefix below mega. The capital B is the byte, the small b the bit, and the difference between them is eightfold. The binary prefixes are written with an «i» and a capital first letter: KiB, MiB, GiB.
In Russian the word for byte does not change in the plural after a numeral: one writes 512 byte, not the plural form. The symbol B never takes a case ending either — it is a symbol and not an abbreviation.
06 · Neighbouring units
The byte stands upon the bit and lives in groups: a nibble of four bits corresponds to one hexadecimal digit, two bytes to a short number, four to an address, eight to a long one. The machine word is a whole number of bytes, and their number tells the architecture.
Four bytes give four thousand million states — that was enough for all the addresses of IPv4, and that is exactly why they ran out.
07 · Historical section
Bytes that were not eight bits
On the machine where the word was born, the length of a byte was set by the instruction: from one to eight bits. Fixing it at eight came later, with System/360.
It was enough for capital Latin letters, digits and signs — there were no small letters in these systems at all. A word of 1 bit was divided into six such characters.
The original encoding was seven-bit: 128 characters. The eighth digit went to a parity check, and only later became part of the code, giving room for national alphabets.
Machines with a 36-bit word sometimes divided it into four parts of nine bits. The System/360 standard gradually drove out all the variants.
Eight bits were chosen not by physics but by one company’s commercial decision
Before 1964 the byte was anything at all: six bits at CDC, seven in teletypes, nine on machines with a 36-bit word, an arbitrary length on the IBM Stretch. Choosing a size for System/360, IBM settled on eight — a power of two, fitting into two hexadecimal digits, enough for the Latin alphabet with its control codes and with room to spare for national characters. The success of the line made eight the de facto standard, and the IEC fixed it only in 2008, forty-four years later. Formally the standards have a separate word for exactly eight bits — the octet — and network specifications write precisely that, to rule out any ambiguity.
Catalogue · units of measurement
A passport for every quantity
Seven SI base units, twenty-two derived ones with their own names, and non-system units alongside them. Each has its own sheet: definition, dimension, conversion, instruments, rules of writing. In 14 languages.