Announcement

Collapse
No announcement yet.

Why do Xeons suck so much?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Why do Xeons suck so much?

    So, I'm specing out a new MPI cluster here at work, and in doing so discovered that P4 Xeons are truly abominable. The highest-end version available uses a shared 533MHz (4 * 133MHz) front side bus, compared to an 800MHz bus on all recent P4 systems. That equates to 266MHz of bus bandwidth per processor when each processor is using the bus at saturation (which tends to happen when you have a CPU bound mesoscale atmospheric modelling program working with large data sets)

    So, I fail to understand, why has Intel managed to make such monumental strides on the desktop P4 side (which now support dual channel DDR400 memory controllers) yet allowed Xeons to slip by the wayside?

    Honestly, I see them completely losing their tenuous grasp on the low end server space to AMD in the near future.
    45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B0
    45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B1
    [ redacted ]

  • #2
    one word. "profitability"

    the xeons WERE the uber coolest thing of their time....

    but more profit can be made from the home computer users. so where are they going to spent most of their research cash?

    home pc based processors. more specificaly, Mobil processors, which are the fastest growing markets by far
    the fresh prince of 1337

    To learn how to hack; submit your request

    Comment


    • #3
      How do Optitons compare, you are refering to the "off the shelf" markert right? Other custom, or expesive beyond belief stuff is way better for clusters and SMP
      "Never Underestimate the Power of Stupid People in Large Groups"

      Comment


      • #4
        Originally posted by hackajar
        How do Optitons compare
        If you're referring to Opterons, the only quote we received was for a 16 processor Opteron 242 cluster, which ran into the $12,000 range. For that price we could purchase at least 4 additional computational nodes given the quote below.

        you are refering to the "off the shelf" markert right?
        No, I'm not referring to the "off the shelf" market as we are forced to pay overhead on parts, negating any cost benefit in building it ourselves (we built our last cluster from parts and my boss is unwilling to pay overhead again)

        Our current quote is approximately $8k for the following setup...

        Master Node:
        Dual 2GHz Athlon MP 2400+
        2GB DDR SDRAM
        8 * 200GB WD 2MB cache drives (= ~1.4TB RAID 5)
        3Ware 8-channel ATA133 RAID controller

        Computational Nodes (x 7)
        Dual 2GHz Athlon MP 2400+
        1GB DDR SDRAM

        + Netgear 8 port 10/100 switch

        Other custom, or expesive beyond belief stuff is way better for clusters and SMP
        To be sure... there's a number of Origin 2100s littering our server room. Everyone in my group has a Blade 1000/2000 or an Ultra 60 for their workstation. However, after seeing the comparitive performance of our first cluster, no one is interested in spending their grant money on a high end workstation; they want a cluster that offers them the most power for their money.
        45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B0
        45 5F E1 04 22 CA 29 C4 93 3F 95 05 2B 79 2A B1
        [ redacted ]

        Comment

        Working...
        X