Results 1 to 3 of 3

Thread: What CPU to use for CAS@Home, and why

Threaded View

Previous Post Previous Post   Next Post Next Post
  1. #1
    Join Date
    Sep 2010
    Location
    Leiden, the Netherlands
    Posts
    4,384

    What CPU to use for CAS@Home, and why

    There's two ways of looking into this problem. One is to use the data of the project itself.

    The top-5 all-time sees two Authentic AMD QEMU Virtual CPUs, version 1.2.0 at the top (36(!) cores each) followed by a dual Xeon L5335, a Xeon E5520 (8 cores/threads each) and a C2Q-8300 (4 cores).
    The top-5 RAC sees a i7-4770K on top (8 threads), followed by a Xeon L5639 (24 threads), a Xeon L5520 (16 threads), a i7-2600 (8 threads) and a Xeon E5620 (16 threads). Highest AMD CPU is a FX-8350 at #12.

    The message seems to be: bring as many cores as you can bring, the faster the better. Remarkable that the i7s with 'only' 8 threads manage to keep up with the multi-multi thread boys, but at $250-350 you consumer get at least some value for money (much money that is).

    Wuprop gives me an extra problem: the data looks f**ed up. It looks like a floating-point error. I guess that -broadly speaking- US and Euopean data have been mixed, disregarding the different use of comma's and periods by those users. I -Europe- write 1.023,45 credits, you write 1,023.45 credits and WuProp adds it up to 1,024,447.345, as it looks to me. So I'm going to postulate that all those mega values ought to be chopped down to around the four left digits credits per day per core, to be done by slicing off all extra digits that appear from the right. Crude, but workable.

    How do the quaddies perform here?
    Scores per core per day
    Phenom II X4-980: 6,497,837.6 (actually X4-970 data) -> chop! = 649.7 per core per day, on 32-bit Windows
    Llano A8-3820: 57,153,600.0 -> chop! = 571.5 per core per day, on 32-bit Windows
    Llano A8-3870K: 1,212.2 For once I'm believing a value. Done on 32-bit Linux, by the way, or so they say.
    FX-4100: 49,737,600.0 -> chop! = 497.4 per core per day, on 32-bit Windows? Who's using Win32 on an FX?
    FX-4170: na
    FX-4300: na
    FX-4350: na
    Trinity A10-5700: 611.2 Also done on a 32-bit OS, this time Linux. I'm not trusting this anymore either...
    Trinity A10-5800K: na
    Richland A10-6700: na
    Richland A10-6800K: na

    Best performing AMD CPU seems to be the FX-8150 that reaches 123,465,600.0 -> chop! = 1,234.6 credits per core per day (I hope! A real minefield this table...)
    Best performing Intel CPU -as I see it- is the Xeon E5-2600 v2 E5-2695 v2, whose 24 hyperthreaded cores reach 553.0 per thread per day = 13,272 credits per CPU per day.
    It is, however, ten times as expensive as a socket 1155/1150 i7. Performance at a price, as the marketing department of Intel keeps telling us.
    Last edited by Dirk Broer; 01-18-2014 at 10:26 AM. Reason: ten times as expensive as a socket 1155/1150 i7


Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •