On Wed, Nov 16, 2005 at 02:26:12PM -0600, Nix, Robert P. wrote: >> Actually, in this case 32 bit and 64 bit refer not to the natural >> word size, but to the address size. i.e. a 32 bit machine uses a 32 >> bit address for memory, and cannot address as much memory as a 64 >> bit machine, using 64 bit addresses, can. >> >> In the case of the IBM mainframe computers, the natural word used to >> be 32 bits, but they were 31 bit machines, as the top bit of the >> word was reserved for something other than an address bit. The new >> mainframes are 64 bit machines, using all 64 bits of an address to >> address memory. They're still 32 bit words though; they just happen >> now to use two of them for an address. There was also an earlier era (late 1970s and before) in which these machines used 32-bit words but only 24-bit addresses, reserving an entire byte of each word for other purposes. The transition from 24-bit addressing to 31-bit addressing was -- messy? a massive kludge? -- since for reasons of upward compatibility IBM ended up providing support for a while (possibly still??) for two addressing modes. There's probably a lesson here for future architects, something along the same lines as not saying in public that 640K ought to be enough for anyone. :-)? >> >> (Actually, the smallest thing the mainframe would deal with for a >> long time (early 1980's forward) has been 64 bits, or a double-word. >> The main change has been the use of larger addresses.) >> -- blm
