January 19, 2026
NaN but make it drama
Kahan on the 8087 and designing Intel's floating point (2016) [video]
The math legend behind Intel’s number chip sparks nostalgia and nerd fights
TLDR: William Kahan explains how he helped design Intel’s floating‑point math for the 8087, the foundation of today’s number handling. Comments split between praising precision (80‑bit, NaNs) and blaming it for complexity and speed, with nostalgic co‑processor tales and warnings to never use floats for money.
The internet’s math nerds are buzzing over a throwback clip of William Kahan—yes, the Turing Award winner—explaining how he helped Intel build its floating‑point math for the 8087 co‑processor, later folded into the 486. Commenters are treating Kahan like the numbers whisperer, crediting him with shaping the IEEE 754 rules that keep computers from turning 1.10 into chaos.
Strong opinions? Oh, they’re spicy. Old‑school engineers are glowing about the beige‑box era (“I saved for months to buy the 387!”), while performance die‑hards grumble that Kahan’s precision features—like special values for “not a number” (NaN) and infinity—made math slower and more complicated. One camp cheers the 80‑bit “extra precision” of the x87 as genius for accuracy, the other calls it a trap that broke code when later chips favored simpler 64‑bit math.
Drama flares over money math: vets scream “never use float for currency,” and a few brave souls insist the real villains are lazy APIs, not Kahan. Memes fly: “Thanks Kahan, my bank balance isn’t NaN,” and “Pentium FDIV flashback” posts link the infamous Intel division bug as proof that correct math matters (remember this?). The vibe: reverence, rivalry, and retro flexing—watch the clip, pick a side, and prepare for precision vs. speed cage match.
Key Points
- •William Kahan consulted for Intel to design a new floating-point system.
- •The floating-point design was first incorporated into the Intel 8087 coprocessor.
- •The design was refined for Intel’s 80287 and 80387 chips.
- •The refined floating-point functionality was integrated into the Intel 80486 processor.
- •The 80486DX denotes the 80486 version with an integrated floating-point unit.