SRAM vs Cache

Discussion in 'VHDL' started by Michael, Sep 18, 2003.

  1. Michael

    Michael Guest

    Hi all!

    What's the difference in "gates density" between a typical
    CPU L1 cache and on chip static RAM?

    I mean.. how many logic gates are necessary for e.g. 4 KB
    of 80486-style L1 cache, and how many for embedded SRAM?

    I guess that for the latter it's no less than 65536 gates
    (please feel free to provide an exact value!), but for the
    L1 cache of equivalent size, how many gates are necessary?
    (with gates I mean basic NAND gates).

    Thank you!
    Mike
     
    Michael, Sep 18, 2003
    #1
    1. Advertisements

  2. Michael

    Mario Trams Guest

    Michael wrote:

    Hello Michael,
    None, actually. CPU cache is usually nothing more than "normal"
    static RAM. Some CPUs might use dual-ported RAM in order
    to parallelize the interfaces between CPU and cache and
    external bus and cache, but I guess that is out of scope of
    your question.
    Because cache is SRAM, there is no difference. However,
    memories are usually not built out of gates, but directly
    out of transistors. Of course, they represent boolean
    equations, finally.
    The common transistor count per SRAM-cell is 6. So for
    4KB, you need 4*1024*8*6 = 196608 transistors. Of course,
    there are a few more ones needed for decoding logic etc.

    If you want to think in NAND-Gates, such one is built out of
    4 transistors normally (assuming a push-pull stage). So your
    4KB RAM needs an equivalent of around 50000 gates. But again,
    SRAMs are not built from gates as such!
    Dito.

    Regards,
    Mario
     
    Mario Trams, Sep 19, 2003
    #2
    1. Advertisements

  3. Michael

    Uncle Noah Guest

    VERY WRONG. Caches incorporate special "tag comparison" logic. They
    are larger than their SRAM counterparts and more power expensive. For
    power it is:
    ROM < SRAM (scratchpad) < cache.
    What can i say about that? Then we should you caches as often as
    possible. No wonder why 512KB of L2 cache is so expensive...

    Uncle "The G.B. Man" Noah
     
    Uncle Noah, Sep 20, 2003
    #3
  4. Michael

    Mario Trams Guest

    Of course, that is an important management aspect of a cache
    as a whole (and I do know very well how caches are working :).
    To justify myself, in my post I referred just to the actual data
    cache because it appeared to be a sufficient answer to the
    original question.
    Dito. Again, this refers just to the actual data storage excluding
    management aspects.

    Regards,
    Mario

    --
    ----------------------------------------------------------------------
    Digital Force / Mario Trams -chemnitz.de

    Chemnitz University of Technology http://www.tu-chemnitz.de/~mtr
    Dept. of Computer Science Tel.: (+49) 371 531 1660
    Chair of Computer Architecture Fax.: (+49) 371 531 1818
    ----------------------------------------------------------------------
     
    Mario Trams, Sep 21, 2003
    #4
  5. Michael

    Mario Trams Guest

    ^^^^^^^^^^ should read "data storage memory of the cache"

    Mario
     
    Mario Trams, Sep 22, 2003
    #5
    1. Advertisements

Ask a Question

Want to reply to this thread or ask your own question?

You'll need to choose a username for the site, which only take a couple of moments (here). After that, you can post your question and our members will help you out.