The 680x0/ix87 Extended formats keep the 8087's explicit leading significand bit 'for historical reasons'; the 8087 features it served were 'deemed too arcane to include in IEEE 754'

ieee-754intel-8087floating-pointextended-precisionwilliam-kahanhistory-of-computing

The 8087-to-IEEE-754 lineage was selective engineering judgment, not a wholesale copy of the chip's arithmetic into the standard. Kahan's own technical status report on IEEE 754 identifies one visible hardware residue of that selectivity: the 80-bit Extended formats of the 680x0 and ix87 chip families carry an explicit (non-implicit) leading significand bit "for historical reasons" — a direct carryover from the Intel 8087, where it let the chip suppress normalizing subnormal numbers advantageously in certain scalar products within matrix computations. The standard itself went the other way: IEEE 754's Single and Double formats use an implicit leading bit, and its optional Double-Extended format leaves the encoding to the implementation rather than specifying the 8087's.

Kahan is explicit about which way the filter cut: "this and other features of the 8087 were later deemed too arcane to include in IEEE 754, and have atrophied." The normalization-suppression capability the explicit bit served was left out of the standard, not carried into it; what survives is the bit itself, fossilized in the chips' extended formats. The standard absorbed the 8087's arithmetic reasoning and its general shape, but individual implementation choices were filtered through the committee's judgment about what a general, hardware-independent standard should specify — not copied wholesale from one vendor's chip.