The first FPU I know of was the AMD AM9511 Arithmetic Processor, circa 1978. It had a stack-based architecture, working with 16-bit and 32-bit fixed point, and 32 bit floating point (24 bits for mantissa, 7 bits for exponent). Very slow for today FPU standards...

Other data:
24 pin-dip
+5Vcc, +12Vdd
8 bit data bus
3 MHz max frequency
32-bit FP operations: FADD, FSUB, FMUL, FDIV, SQRT, SIN, COS, TAN, ASIN, ACOS, ATAN, LOG, LN, EXP, PWR.
Many data and stack manipulation operations

Regards,
Txip.