SNB L1D bank conflicts?

By: David Kanter (, February 7, 2012 1:57 pm
Room: Moderated Discussions
One of the things that struck me as odd about SNB is that the banking on the L1D cache doesn't seem all that robust.

If I remember correctly, there are 8 banks, which each drive 16B/cycle. That means if you are sustaining 48B/cycle, you are occupying 6/8 banks. In contrast, I believe Bulldozer has 16 banks.

For regular and array like accesses, this shouldn't be a problem at all. However, for more complicated access patterns, conflicts may emerge.

Does anyone have data or anecdotal experience with these issues?

 Next Post in Thread >
TopicPosted ByDate
SNB L1D bank conflicts?David Kanter2012/02/07 01:57 PM
  SNB L1D bank conflicts?EduardoS2012/02/07 02:01 PM
    SNB L1D bank conflicts?David Kanter2012/02/07 02:39 PM
      Can OoO help?Mark Roulo2012/02/07 05:06 PM
      SNB L1D bank conflicts?Eric Bron2012/02/08 03:59 AM
      SNB L1D bank conflicts?Eric Bron2012/02/08 04:21 AM
    SNB L1D bank conflicts?Michael S2012/02/07 02:54 PM
  SNB L1D bank conflicts?Linus Torvalds2012/02/07 02:58 PM
    Usefulness of bank skewing?Paul A. Clayton2012/02/07 04:48 PM
      Usefulness of bank skewing?Linus Torvalds2012/02/07 05:47 PM
        Another idea bites the dust :-)Paul A. Clayton2012/02/08 05:42 PM
  SNB L1D bank conflicts?blaine2012/02/07 10:49 PM
  SNB L1D bank conflicts?Michael S2012/02/08 01:53 AM
    SNB L1D bank conflicts?EduardoS2012/02/08 03:29 PM
    SNB L1D bank conflicts?Tom2012/02/11 05:57 PM
  SNB L1D bank conflicts?Walter L.2012/02/14 07:26 PM
Reply to this Topic
Body: No Text
How do you spell green?