
Intel 8087
A 1980 floating-point companion chip for the 8086 and 8088 CPUs
The Intel 8087 is a floating-point coprocessor that Intel announced in 1980 to work alongside its 8086 and 8088 microprocessors. It handled maths that the main CPU could not do quickly, and its number formats shaped the IEEE 754 standard used across computing today.
- Announced in 1980 as a floating-point partner for the 8086 and 8088
- 65,000 transistors on a 4.5 µm process, later shrunk to 3 µm
- About 50,000 FLOPS at roughly 2.4 W
- Its formats became the basis for IEEE 754-1985
- Sold in 4.77 (5), 8 and 10 MHz speed grades
What it is
The 8087 is a companion processor built to handle decimal and real-number arithmetic for Intel's 8086 and 8088 CPUs. It covers addition, subtraction, multiplication, division and square roots, along with exponential, logarithmic and trigonometric routines. Speed-ups varied widely by workload, from roughly 20 per cent to more than 500 per cent. Throughput reached about 50,000 FLOPS while drawing close to 2.4 watts. Intel announced the part in 1980, and IBM later sold it too. Its number formats fed into the IEEE 754-1985 floating-point standard, which remains the reference for how computers represent real numbers.
How it came about
Bill Pohlman, the engineering manager behind the 8086, planned for a maths companion from the start and won approval for the project in 1977. Bruce Ravenel took the architect role, with John Palmer joining as co-architect and mathematician. Their work drew on William Kahan's writing about floating point. The ambitious design met a cool response in Santa Clara, so implementation moved to Intel Israel under Rafi Nave. Palmer, Ravenel and Nave held patents on the design, while Robert Koehler and John Bayliss patented the method for routing selected instruction patterns to the coprocessor.
Inside the chip
Numbers sit in an eight-level stack of 80-bit registers, with st0 at the top. Instructions push, compute and pop values, and the FXCH swap makes the stack behave like seven addressable registers plus an accumulator. The longest real format carries a 64-bit mantissa and a 16-bit exponent. Data types include 32-bit and 64-bit floats, an 80-bit extended temporary form, packed decimal and several integer widths. Around 60 new instructions arrive, most with mnemonics starting with F. There is no hardware multiplier, so CORDIC handles those tasks. Basic adds and subtracts can pass 100 cycles, and some instructions run beyond 1,000.
How it was made
Intel built the chip with 65,000 transistors on a 4.5 µm depletion-load HMOS process, later reduced to 3 µm. Microcode lived in a multi-level ROM that packed two bits per location by choosing among four current levels, saving silicon area. Yields stayed low and the part ran hot, so Intel used a costlier ceramic 40-pin DIP for heat spreading. AMD and Cyrix also manufactured it.
Where it went
The chip paired with 8086 and 8088 systems, and sales rose sharply after the 1981 IBM PC motherboard included a socket for it. Programs had to be written to use the special instructions; at run time software could probe for the coprocessor and switch to it. Where no coprocessor was fitted, the same maths ran in software, or the whole unit was emulated for numerical accuracy. Because the CPU kept control of the bus and the coprocessor only borrowed it briefly, integer work and floating-point work could proceed at the same time.
Versions and family
Speed grades were 4.77 (5), 8 and 10 MHz. Intel had earlier sold the 8231 arithmetic unit and 8232 floating-point processor for 8080-class systems over an 8-bit bus. Later x87 companions served the 80186, 80286, 80386, 80386SX and 80486SX. From the 80486DX onward the floating-point unit sat inside the main processor. A 486SX could still take an 80487, which was really a 486DX with an extra pin that switched off the built-in CPU.
Why it matters
The 8087 set the shape of x87 floating point and became the reference for IEEE 754-1985. Its stack-based register model, extended precision and instruction naming carried into every later Intel maths unit. The design also showed how a coprocessor could share a bus with a CPU and run in parallel, a pattern that influenced later system design. Its formats for infinity, in both affine and projective forms, entered the draft standard, though projective closure was dropped from the final 1985 text.
Problems and reception
The aggressive design drew a cool reception inside Intel before implementation moved to Israel. Manufacturing proved difficult, with low yields and high heat output that forced a ceramic package. A crash risk existed if the coprocessor failed to decode an instruction, since it sat directly on the address and data buses. Later Intel coprocessors avoided this by receiving instructions through CPU I/O ports instead, at some cost in speed.
Questions people ask about Intel 8087
What did the 8087 actually do?
It sat beside the main CPU and handled real-number arithmetic: add, subtract, multiply, divide, square root, plus exponential, logarithmic and trigonometric routines. Programs issued special instructions that the coprocessor picked up from the shared bus while the CPU carried on with other work.
How much faster did it make things?
Gains depended on the program, ranging from about 20 per cent to more than 500 per cent. Raw throughput reached roughly 50,000 FLOPS while consuming close to 2.4 watts.
Why does it matter for how computers handle numbers today?
Intel designed its formats with a future standard in mind, and the 8087 became the basis for IEEE 754-1985. That standard defines how floating-point values, rounding and infinity are represented, and it is still used across the industry.
What happened to the coprocessor idea later?
Companion chips continued for the 80186 through the 80486SX. Beginning with the 80486DX, the floating-point unit was built into the main x86 processor, so a separate part was no longer needed.
Good reading and sources
The facts on this page come from the Wikipedia article "Intel 8087", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryChip rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.
Like Intel 8087
All CPUs and microprocessors
🇮🇱 IsraelIntel 8088
A 16-bit Intel microprocessor with an 8-bit external data bus, used in the original IBM PC.CPU · x86 · 1979
🇺🇸 United StatesIntel 80286
16-bit x86 CPU with protected mode and 24-bit addressing, launched in 1982CPU · x86 · 1982
🇺🇸 United StatesIntel 80186
16-bit microprocessor with integrated peripherals for embedded systemsCPU · x86 · 1982