[cesarb]

> imagine a 16-bit floating point format (P&H call this the "Nvidia format", but I can't find that documented anywhere but there): 1-bit sign, 5-bit (biased) exponent, 10-bit mantissa

According to Wikipedia (https://en.wikipedia.org/wiki/Half-precision_floating-point_...), this is the IEEE 754 standard binary16 format.

[commandlinefan]

Sort of - I skipped over subnormal numbers.