On 26 Oct 2009 at 19:37, Dean Kukral wrote:

> There is a little confusion here that I think that I can clear up.  A
> 64 bit system - "a 64 bit architecture" - refers to the word length
> internal to the cpu.  It generally represents the number of memory
> locations that can be addressed in one word, two to the sixty-fourth
> power.  I believe that all home computer cpu's have had a  64 bit
> architecture for a good number of  years.

  There are actually THREE numbers of interest here:

1.  The size of integer data registers in the CPU.  When people 
talk about a given CPU as being "X bit", this is almost always 
the number they are actually referring to.  This is still 32 
bits for most Intel and compatible processors unless they are 
designated "64 bit" which, although becoming more and more 
common, is still a minority.

2.  The size of address registers in the CPU.  This used to 
commonly exceed the size of the integer data registers:  most 8-
bit CPUs used 16-bit addresses and the 16-bit PC CPUs (8086 
through 80286) used 24-bit addresses.  32-bit data registers 
with 32-bit addressing has kind of been a sweet spot for almost 
20 years.

3.  The width of the physical memory address bus (collection of 
pins) on the CPU, which is almost never wider than the address 
register size.  This determines the actual maximum addressable 
physical memory.  For instance, the 80386 SX and DX processors 
both used 32-bit address registers, but the SX had only 24 
physical address pins, limiting it to 16MB of physical RAM (but 
making it easy to adapt tp motherboard designs intended for 
80286 CPUs...).

  Windows 3.x was a 16-bit OS.  Most of the machines that ever 
ran it had 32-bit CPUs (80386 or better), and a popular 
optimization trick was to use the 32-bit data registers (which 
required at least an 80386) for arithmetic while sticking with 
the 24-bit address architecture from the 80286.  The Pentium Pro 
CPU had an optimized 32-bit core, but it's 16-bit subsystem 
lacked performance and so it did poorly with Windows customers.  
Intel corrected that in the Pentium II, and Microsoft finally 
released Windows 95 and NT 4,
  Now we have CPUs available which provide 64-bit data and 
address registers, and since XP Microsoft has been shipping OS 
versions which support and take advantage of those.  A 64-bit OS 
version won't run on a CPU that isn't also 64-bit.  A 64-bit OS 
knows how to manage RAM that could have physical addresses 
bigger than 32 bits, so is a prerequisite for systems using more 
than 4GB of total address space.  (But the only 64-bit CPU I 
currently own is on a motherboard that is limited to 2GB of RAM, 
so CPU and OS support while required is not sufficient.)

  Windows 95 was able to support 16-bit applications by running 
them in one or more virtual 16-bit environments, and the 64-bit 
versions of Windows are able to do similar (better, actually) 
for 32-bit applications which are still the vast majority.  That 
trick doesn't really work for the OS itself, though, or for 
drivers which need to act as OS components.  If you are buying 
or building a new system out of pretty standard components, that 
shouldn't be an issue, but if you need to support some oddball 
or legacy peripherals, lack of 64-bit driver support may rule 
out a 64-bit OS version.  But you may still find the price/speed 
of a 64-bit CPU and matching motherboard competitive with 32-bit 
versions, and the 32-bit OS and drivers etc will work on them 
just fine.

David Gillett
 

            Do you want to signoff PCBUILD or just change to
                    Digest mode - visit our web site:
                   http://freepctech.com/pcbuild.shtml