Comment by markus_zhang

I checked the documentation and looks like even the 4K version supports single precision floating point numbers (32-bit). Not sure how they managed to do it in 8-bit machines but I guess they just had to use more registers/memory for each of them.