actually i would like to correct myself. all the FPU (ST) registers are 80-bit but the instructions are processed at 64-bits. however, the fld and fstp conversion routines do operate at the same speed and other FPU computation routines which process ST registers "fadd" or "fmul" operate at the same regardless of what the original variable data type was. i.e. 32-bit or 64-bit real, or 16-bit or 32-bit integer. there is no difference in speed with regards to FPU instructions. maybe on a 286.

Jenova :

The problem is the data bus transfer which takes 2 cycles for a 64bit float and only one for a 32bit float if using a 32bit wide data bus.
[And I ask myself the question are we using 32bit wide data buses in our PCs??]

I do not know that INTEL FPU were using 64bits float internally, if so why are they 32bit CPU and not 64bits?

(I asked that question a day: How do I know what bit number is a CPU ? And never got replies.)

-* Sounds, music and story makes the difference between good and great games *-

ok. first of all the CPU and the FPU are two completely different processors. the FPU processor is a dedicated processor for doing floating point instructions. yes the Intel 80386 architecture which includes Pentium, Pentium II, and Pentium III use a 32-bit based processor. however, this does not require the FPU to use 32-bit registers. and the FPU does not use 32-bit registers, it uses 80-bit registers. 32-bit computations really present rounding errors so that is why FPU instructions perform computations at the 64-bit precision. the ONLY REAL difference between using floats and doubles is the size of memory used, and doubles are not used with SIMD instructions as stated above.