Bits to Bytes (bit to B)

One bit is the information content of a single binary digit — equivalently, the Shannon entropy of an outcome with probability ½, such as a fair coin flip whose result is then revealed. Formally, for a discrete random variable X with probability mass function p(x), the Shannon entropy is H(X) = −Σ p(x) log₂ p(x) bits. The choice of log base 2 fixes the unit as the bit; log base e gives the nat (natural unit, ≈ 1.443 bits); log base 10 gives the hartley or ban (≈ 3.322 bits). The bit is the foundational unit in coding theory, channel-capacity calculations (the Shannon–Hartley theorem expresses maximum reliable channel capacity in bits per second as a function of bandwidth and signal-to-noise ratio), in cryptography (key lengths and hash output sizes are denominated in bits), and in computer-architecture word widths (32-bit and 64-bit address spaces, 8-bit and 16-bit microcontroller cores). The bit is conceptually distinct from the binary digit considered as a written symbol — the digit "0" or "1" on a page carries 1 bit of information only when the two outcomes are equally likely; a coin biased 90/10 has Shannon entropy of just 0.469 bits per toss.

The bit is the fundamental unit of information, and unlike the byte it has a specific intellectual paternity rather than a hardware-engineering one. The word "bit" — a contraction of "binary digit" — was coined by the American statistician John W. Tukey in a Bell Labs internal memorandum dated 9 January 1947, in which Tukey rejected the alternative coinages "binit" and "bigit" as cumbersome. Tukey was at the time consulting at Bell Labs and would go on to coin "software" in 1958 and to co-develop the Cooley–Tukey FFT in 1965; the bit was an early entry in a remarkable lifetime of terminology. The unit's formal mathematical definition came eighteen months later, when Claude E. Shannon published "A Mathematical Theory of Communication" in the Bell System Technical Journal across the July and October 1948 issues — the founding document of information theory. Shannon defined the information content of an outcome with probability p as −log₂ p, measured in bits, and the entropy of a discrete random variable as the expectation H = −Σ p log₂ p, also in bits. Shannon's 1948 paper rested on two earlier Bell Labs predecessors. Harry Nyquist's 1924 paper "Certain Factors Affecting Telegraph Speed" introduced the relationship between channel bandwidth and signalling rate; Ralph Hartley's 1928 paper "Transmission of Information" introduced a logarithmic measure of information content using log base 10 (the unit later renamed the hartley or ban). Shannon chose log base 2 because it aligned naturally with binary signalling and with the relay-and-switching circuits he had analysed in his 1937 MIT master's thesis "A Symbolic Analysis of Relay and Switching Circuits" — the thesis, supervised by Vannevar Bush, that mapped Boolean algebra onto electrical-relay networks and is widely regarded as the most important master's thesis of the twentieth century. The IEC standardised the bit's symbol as the lowercase "b" in IEC 60027-2 to disambiguate from the byte's uppercase "B".

Networking is the bit's defining domain in everyday computing — every transmission rate published by an ISP, telecommunications carrier, Wi-Fi alliance, or fibre-backbone operator is denominated in bits per second by long industry convention. The 8:1 ratio between bits-for-networking and bytes-for-storage is the single most consequential unit-system boundary in consumer technology: a "1 Gbps" gigabit Ethernet link delivers 125 MB/s of file-transfer throughput; a "200 Mbps" residential broadband connection delivers 25 MB/s; a "5G NR" sub-6 GHz radio link sustaining 1 Gbps delivers 125 MB/s; a "Wi-Fi 6E AX5400" router advertises 5,400 Mbps aggregate radio capacity but per-device usable goes through the ÷8 conversion. The convention is universal across the IEEE 802.3 (Ethernet), 802.11 (Wi-Fi), and 3GPP (cellular) standards documents, in DOCSIS cable specifications, and in undersea-cable capacity quotes (the MAREA Atlantic cable launched in 2018 at 160 Tbps, equivalent to 20 TB/s of byte-throughput). Cryptography denominates key strength, hash output, and signature-scheme parameters entirely in bits. AES (FIPS 197) is specified at 128, 192, and 256-bit key lengths; SHA-2 produces 224, 256, 384, or 512-bit digests; RSA key lengths run 2048, 3072, 4096-bit; elliptic-curve cryptography Curve25519 operates with 256-bit private keys; the post-quantum standards selected in NIST PQC Round 4 (CRYSTALS-Kyber, CRYSTALS-Dilithium) specify parameter sets in bits-of-classical-security and quantum-security. The 256-bit baseline has become the universal "AES-256, SHA-256, ECDSA-256" reference for symmetric-equivalent strength against both classical and near-term quantum adversaries. Display and imaging use bits to describe per-channel colour depth and high-dynamic-range gradation. Standard sRGB displays are 8-bit per channel (24-bit total RGB, 16.7 million colours); HDR10 and HDR10+ specify 10-bit per channel (1.07 billion colours, the Rec.2020 colour space); Dolby Vision specifies 12-bit per channel (68.7 billion colours); cinema-grade colour grading and DCP delivery work in 16-bit per channel. The visible banding artefacts in 8-bit gradients on HDR-capable televisions are the consumer-visible reason 10-bit panels are now standard on premium displays. Audio bit depth determines dynamic range. CD audio is 16-bit / 44.1 kHz (theoretical 96 dB SNR, ~98 dB after dithering); high-resolution streaming on Apple Music, Tidal, and Qobuz delivers 24-bit / 96–192 kHz lossless (theoretical 144 dB SNR); professional studio recording and mastering work in 32-bit float for unlimited internal headroom. Telephone-grade voice is 8-bit μ-law or A-law companded at 8 kHz, the legacy bit-rate of the public switched telephone network. Computer architecture: 8-bit microcontrollers (AVR, PIC, 8051) dominate cost-sensitive embedded designs; 32-bit ARM Cortex-M cores dominate IoT and industrial control; 64-bit ARM, x86-64, and RISC-V cores dominate phones, laptops, servers, and supercomputers. The CPU bit-width describes both the natural integer register size and the addressable memory space (a 32-bit address space tops out at 4 GiB, the practical RAM ceiling that drove the 32→64-bit consumer transition between 2003 and 2010).

One byte equals exactly 8 bits, encoding 2⁸ = 256 distinct values (0–255 unsigned, or −128 to +127 in two's-complement signed representation). The byte is the smallest individually-addressable unit of memory in essentially all modern computer architectures, and is the unit in which file sizes, memory capacities, and storage media capacities are reported. The IEC formal name for the 8-bit byte is the "octet" (IEC 60027-2:2005, IEC 80000-13:2008), used in international standards documents and in networking RFCs to disambiguate from historical machines where "byte" meant other widths. ASCII (ANSI X3.4-1968) encodes a single character in 7 bits within an 8-bit byte; UTF-8 (RFC 3629, 2003) encodes the full Unicode code-point repertoire in 1 to 4 bytes per code point, with the ASCII subset occupying 1 byte for backward compatibility. By IEC 80000-13, the symbol B denotes the byte and is distinguished from the bit (symbol b) by capitalisation alone — a typographic distinction that survives intact through internet-speed advertising in Mbps (megabits per second) and storage capacities in MB (megabytes).

The byte was coined in June 1956 by Werner Buchholz, an IBM engineer working on the IBM Stretch (7030), the company's first transistorised supercomputer project. Buchholz needed a name for an addressable group of bits smaller than a machine word, and proposed "byte" — deliberately respelled from "bite" so a single typographical slip could not collapse the new unit into "bit". The respelling was not cosmetic: in early documentation handwritten and typed in mixed lower- and upper-case, "bit" and "bite" were dangerously similar, and the "y" gave the word a unique fingerprint at a glance. Through the late 1950s the byte was variable-width — Stretch's byte ranged from 1 to 8 bits depending on the operation, and contemporary machines such as the IBM 7090 used 6-bit characters. The 8-bit byte was standardised by the IBM System/360 architecture announced on 7 April 1964, which addressed memory in 8-bit bytes and grouped four bytes into a 32-bit word. System/360's commercial success made the 8-bit byte the de facto industry standard, and ANSI X3.4-1968 (ASCII) ratified the 7-bit character within an 8-bit byte for North American computing. The IEC formalised the 8-bit byte as the "octet" in 1998 (IEC 60027-2) to remove the residual ambiguity in international standards work, although in everyday English-language usage "byte" universally means 8 bits. The byte's defining structural problem appeared once storage capacities scaled past the kilobyte: SI prefixes (kilo, mega, giga) are powers of 1000 by long-standing physics convention, but powers of 1024 are mathematically natural for byte-addressed memory and were used by every operating system from CP/M onwards. The IEC 80000-13 standard published in March 2008 introduced the unambiguous binary prefixes — kibibyte (KiB = 1024 B), mebibyte (MiB = 1024² B), gibibyte (GiB), tebibyte (TiB) — but adoption outside Linux distributions and a handful of standards-conscious vendors has been slow.

The byte is the foundational unit of digital information across every computing system in use. Character encoding is its most universal application: every plain-text file, source-code file, email body, JSON document and HTTP header is denominated in bytes, with ASCII assigning 1 byte per English-language character and UTF-8 using 1–4 bytes depending on the Unicode code point (Latin letters and digits remain 1 byte; Cyrillic and Greek 2 bytes; CJK ideographs 3 bytes; emoji and supplementary planes 4 bytes). File-system sizes, memory allocations, network buffer sizes and CPU cache lines are all reported in bytes or byte multiples. The byte is also where the binary/decimal prefix war originates and propagates. Storage manufacturers (Seagate, Western Digital, SanDisk, Samsung) follow the SI convention literally, advertising 1 TB = 10¹² = 1,000,000,000,000 bytes — the convention that gives the largest possible marketing number. Operating systems historically use binary multiples: Microsoft Windows reports a "1 TB" drive as 931 GB because Windows interprets "GB" as 2³⁰ = 1,073,741,824 bytes, exposing a 7.4% gap that has launched a generation of consumer-support escalations. Apple macOS realigned with the SI convention in OS X 10.6 Snow Leopard (2009), reporting drive capacities in decimal gigabytes that match the manufacturer's marketing. Linux distributions vary by tool: `ls -h` and `df -h` use binary kibibyte/mebibyte/gibibyte semantics with SI-style "K/M/G" symbols, while `ls --si` and the GNU `coreutils` documentation expose the IEC distinction explicitly. Network engineering, by long convention, uses decimal multiples for bandwidth (Mbps in millions of bits per second) — a convention covered in detail under the bit, mbps and gbps entries.