Size of a Word and addressing

Question

I was refreshing myself on memory information and I am confused on the size of a Word. From my understanding, a Word is not a universally defined size, but is a size defined by the specific system (in terms of number of bytes).

According to wikipedia:

Hence, a processor with 32-bit memory addresses can directly access 4 GiB of byte-addressable memory.

Does that mean that a 32-bit processor can address 4,294,967,295 Words? 32-bit windows is limited to 4GB of RAM, but reading over the meaning of a word had me wondering. Does each Word in windows equate to 1 byte? Could the size of a word just be a larger number of bytes and a 32-bit processor be able to address 8GB, 10GB, 12GB or even more memory?

Answer 1

Does that mean that a 32-bit processor can address 4,294,967,295 Words?

It depends on the CPU and on how you look at it.

There are CPUs that can't address anything smaller than a word. 16-bit Texas Instruments Digital Signal Processors are a good example. Their C/C++ char (AKA byte), short and int types are all of the same size, 16 bits. And that is the smallest unit of memory that can be addressed with a unique address (pointer) and that's the machine word at the same time. Since data addresses/pointers are 16-bit on these processors, they can address at most 2¹⁶ 16-bit words in the data memory.

Now, if you go back to x86 CPUs in 32-bit modes of operation, things are a little different. The smallest addressable unit of memory is an 8-bit byte and the largest is a 32-bit word (machine word). Addresses and pointers are 32-bit as well (if we ignore segmentation and page translation). This lets us have 2³² unique memory addresses. And, trivially, with them you can access up to 2³² 8-bit bytes of memory. But how many 32-bit words can you address with 2³² unique addresses? The answer depends on whether you want non-overlapping or overlapping words. You see, the x86 CPU can access 32-bit units of memory at any address, not just at addresses that are multiple of 4 bytes.

You aren't just limited to this on x86:

  0 1 2 3 4 5 6 7  <- address
  \word/  \word/

These all are all valid addresses for 32-bit word accesses on x86:

  0 1 2 3 4 5 6 7  <- address
  \word/  | | | |
    \word/  / | |
      \word/  / |
        \word/  /
          \word/
            ...

So, how many 32-bit words can you address with 2³² unique addresses on x86? If you're talking about all uniquely addressable and overlapping ones, that's 2³² of them. If, OTOH, you're talking about all uniquely addressable and non-overlapping ones, that's 2³⁰ of them.

OTOH, if your 32-bit CPU uses non-32-bit addresses, the total count will be different.

Answer 2

You are confusing word size and byte size, because word size is determined by the processor and byte size is universal.

Without using PAE (Physical Address Extension) a 32-bit processor can only handle 4GB of RAM since there are only 2 ^ 32 = 4,294,967,296 different numbers that can be represented using 32 bits. This is a hardware limitation.

32-bit systems utilizing PAE can address more than 4GB, by mapping processes to page tables. This grants each process access to 4GB of memory. Via supporting PAE, the 32-bit version of Windows Server 2003 Datacenter supports up to 64 GB of RAM on x86-based computers.

Edit

Word size denotes the amount of bits a CPU can process at one time. So 32 bits = 4 bytes. However, word size and address size are often used interchangeably which leads to confusion, since they are indeed different. The Pentium Pro was a 32-bit CPU with a 36-bit wide address bus, allowing 64GB of accessible memory (via PAE.)