The Bit-Width Compendium

The 4-bit spark

The commercial microprocessor begins with the Intel 4004 in 1971 — a 4-bit part designed for a desktop calculator, packing roughly 2,300 transistors and running at 740 kHz. Four bits is barely enough to represent a single decimal digit, and the 4004 addressed program and data memory through separate, narrow spaces. Yet it proved the thesis that would define the next half century: an entire processor could be etched onto a single piece of silicon and sold as a component. The 4-bit generation lived on for decades in the least glamorous, highest-volume corners of electronics — calculators, appliances, toys and controllers — where cost mattered and arithmetic was modest.

The 8-bit home revolution

Eight bits changed everything, because eight bits is exactly one byte — enough to hold a character, a small integer, or a pixel's worth of colour. The Intel 8080, the Zilog Z80 and the gloriously cheap MOS 6502 put real computing into homes, schools and arcades. The 6502 alone powered the Apple II, the Commodore 64, the BBC Micro, the Nintendo Entertainment System and the Atari line. All of them shared the same fundamental shape: 8-bit registers, a 16-bit address bus, and therefore a hard 64 KiB memory ceiling. Programmers learned to love and hate that ceiling in equal measure, inventing bank-switching tricks to reach beyond it and hand-tuned assembly to squeeze cathedral-sized programs into a closet.

This is the era the domain of this very site tips its hat to: the 8-bit machine, where every byte was counted and every cycle was precious.

The 16-bit expansion

Sixteen bits lifted the arithmetic ceiling to 65,535 and, more importantly, opened the door to megabyte-class memory. The Intel 8086 (1978) reached 1 MiB through its segment:offset scheme, and its cost-reduced sibling the 8088 became the heart of the IBM PC. In a different philosophical camp, the Motorola 68000 offered programmers a clean, flat, 32-bit-flavoured model over a 16-bit ALU, and went on to power the original Macintosh, the Amiga, the Atari ST and the Sega Genesis. The 16-bit years were the console and workstation wars, a period when the industry could not agree whether the future was segmented or flat, big-endian or little-endian, CISC or the rising RISC insurgency.

The 32-bit flat-memory revolution

Thirty-two bits retired the segment. A flat, linear 4 GiB address space — vast beyond imagining in 1985 — meant a pointer was just a number again, and operating systems could offer each process the illusion of a private, contiguous memory through paging and virtual memory. The Intel 80386 brought this to the PC and defined the IA-32 platform that dominated for two decades. In parallel, the ARM architecture — born at Acorn in Cambridge — showed that a lean, 32-bit RISC design could be astonishingly power-efficient, a trait that would later make it the default processor of the mobile world. The 32-bit era is where modern software culture was largely set: protected memory, preemptive multitasking, and the C runtime we still argue about.

The 64-bit present

Sixty-four bits removed the last hard wall. The DEC Alpha pioneered mainstream 64-bit computing in 1992, but it was x86-64 — AMD's pragmatic 64-bit extension of the familiar x86, shipped in the Opteron in 2003 — that carried the whole industry across. A 64-bit pointer can name 16 exbibytes of memory, so for the first time the address space stopped being the constraint that forced the next migration. What 64 bits actually bought most users was not gigantic integers but the ability to map more than 4 GiB of RAM per process, stronger cryptography at native word size, and — on x86-64 specifically — twice as many registers. Today the 64-bit word is so thoroughly assumed that 32-bit support is quietly being retired from operating systems and app stores alike.

Beyond the integer: vectors and the future

With the address space no longer forcing the pace, “width” migrated to the data path. SIMD vector units grew from 64-bit MMX to 128-bit SSE, 256-bit AVX and 512-bit AVX-512, each generation packing more parallel lanes for media, scientific computing and machine learning. Meanwhile the open RISC-V instruction set reopened the architecture conversation from scratch, defining clean 32-, 64- and even 128-bit base variants for a world that no longer takes any single vendor's word size for granted. The next widening, when it comes, is likelier to be about vectors, memory tagging and security than about a bigger integer — but the pattern holds: software fills the space it is given, then asks for more.