Yes, sorry your right. Haven't done any 8086 for a long time.
AntiPasta said:
Well actually it's a bit different, the segment register overlaps with the offset to form a 20 bit address from two 16 bit registers.
That used to be the measurement and things have blurred a lot these days.
AntiPasta said:
And your use of the accumulator width as the "bitness" of the CPU raises an interesting point... what about DSPs with their odd, for instance 42 bit accumulator sizes? Are they 42 bit processors? And would that make FPUs 80 bit or more (internal extra precision bits) or less (64 bit as "preferred" data type)?
I suppose the thing is that DSPs and FPUs are not really general purpose CPUs, but I don't see that the principal really breaks down.
DSPs in particular can be any of a wide range of componants, from single chips to multiprocessor boxes. I know some rack DSPs actually contain a couple of 68000s. With DSP units, the number of bits is really more a unit of detail rather than a speed issue.
FPUs still have to break the float down into binary words at some point in the process (at the very least, the mantissa and exponent parts), and it would be this point at which the processor would be classified. It's possible to have an 80 bit FPU with an internal architecture of 8, 16 or 32 bits (or any other number you choose) without affecting the actual results, just the time it takes to get that result.
You could take the argument further and ask if a dual 32 bit core processor is really a 64 bit processor. As I've said, things are a lot more blurred with current tech, and will probably only get more complicated. Look at the way Intel and AMD have changed the speed ratings of their current processors. It's no longer a simply matter of clock speed.