The 8086/8088/80186/80188, 80286 & 80386 all required an external chip for floating point. The 80386DX was the full 32-bit data bus version. The 80386SX was a

The 8086/8088/80186/80188, 80286 & 80386 all required an external chip for floating point. The 80386DX was the full 32-bit data bus version. The 80386SX was a cost reduced version with a 16-bit data bus (akin to the difference between the 8086 & 8088). The 80486DX had an FPU; the 80486SX had a disabled or missing FPU (depending on vintage). You could pair the 80486SX with an 80487 FPU, which was just a full 80486DX in a different pinout that took over from the crippled part. While some Intel doco did start calling them Intel386 & Intel486, they didn't drop the 80 until Pentium. Intel lost a court case trying to prevent clone makers from calling their chips '80486' or whatever because the judge ruled you can't keep your competitor from using a number. Weitek made floating point units that paired to the 386 and 486. Some motherboards supported this, and a small amount of software did (early Autocad). Correcting one down thread, the 486 is more than just a 386+387. There a cache, it's pipelined, and there are some changes to the ISA. 4x the transistors.