Well . cracked 15sec Pi 1M

Discussion in 'AMD x86 CPUs and chipsets' started by m1one, Jul 6, 2010.

  1. randomman

    randomman Member

    Joined:
    Oct 21, 2007
    Messages:
    4,838
    Location:
    Vancouver, BC
    Super Pi is doing 19 loops.

    14/19 = 0.73

    Not that much slower.
     
  2. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,831
    Location:
    Griffin , Brisbane
    No, it's not 19 loops of 1M digits, they're Iterations of a formula, and it takes 19 goes to achieve 1Million Digits of accuracy.. 20 for 2M, 21 for 4M and so on..

    This is the nature of the Gauss-Legendre algorithm that Super-pi uses, details of which, see the help file in super-pi!

    More modern programs don't use this algorithm at all, it's not the fastest now (regardless of code). My guess is in the early 90s it was the fastest way to go about it on hardware at the time, but i'm no Programmer.

    Interestingly, as you'll read in help file. It too 35 MINUTES to compute 1M on a Pentium 90.. Even with Legacy Software like this, it's always good to see how far we've come (albiet have hit a wall with single threaded performance) and THAT's what Super pi is good for
     
    Last edited: Jul 7, 2010
  3. SingAlong

    SingAlong (Banned or Deleted)

    Joined:
    May 30, 2010
    Messages:
    474
    Location:
    Sydney 2035
    Please try it! :D
     
  4. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    5,731
    Location:
    Dark City
    If your looking at whos faster at maths..

    http://gmplib.org/

    Look at the benches. AMD is all over intels i7's. Beating them by 50%. Its all open source and its all able to be optimised if you wish to do it yourself (or intel for that matter).

    This is a library any programmer can use, free of charge and highly reliable. So for scientists, engineers, mathematicians etc this is what they would use if you wanted to find out Pi to a billion decimals. Heck this is what some projects at JPL and NASA use. (non mission critical).

    Shops that roll there own code often buy AMD because its actually outright faster, for them and what they want.

    Still hard core math heads get chips fabricated for what they want to do or use programmable gates. Proberly faster per core by 1000 than anything existing.

    Interesting in next versions you should be able to calculate up to 10^15 digits of pi. You would need 4,000 harddisks just to store that number!

    PS: there are operations that intel CPU's do very fast. And number crunching does not make a better desktop cpu for most users.
     
  5. conroe

    conroe Member

    Joined:
    Apr 19, 2007
    Messages:
    341
    Location:
    Sydney
    Nah mate, you're a long way off. An i5 650 will bury your score with 2 core less.

    So according to your theory, you've got a dud right. Bloody stupid on spending over $330 when a $130 cpu can outgun it.

    Or on the other hand maybe you're just jealous :p
     
  6. randomman

    randomman Member

    Joined:
    Oct 21, 2007
    Messages:
    4,838
    Location:
    Vancouver, BC
    My bad. I always figured it was doing 19 runs as 1 would be completed too quickly.
     
  7. Phido

    Phido Member

    Joined:
    Jun 20, 2003
    Messages:
    5,731
    Location:
    Dark City
    Well modern pi programs will find it in about 1/20th of the time..

    using superpi as a benchmark is like using Whetsonte88 or PCmark or Winbench 95. It does not have a whole bunch of relevance to performance on any modern software. People just use it because its tiny and doesn't take long to use, oh and its been used for the last 10+ years.
     
  8. Psychotik

    Psychotik Member

    Joined:
    Jan 2, 2005
    Messages:
    2,472
    Location:
    127.0.0.1
    And I could OC to 4.1 and probably equal or beat your time. My CPU is 100% stable at this speed so I'm happy.

    Your 650 is not a quad core and therefore USELESS in games such as BC2 that uses one core for SOUND alone.
     
  9. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,831
    Location:
    Griffin , Brisbane

    Hangon, Didnt you just FAIL AMD on super-pi performance?

    which is more important, BFBC2 or Super Pi?
     
  10. pele93

    pele93 Member

    Joined:
    Dec 29, 2009
    Messages:
    504
    Location:
    Canberra
    What a silly question, any enthusiast would know that Super Pi is what determines your e-peen and is vitaly more important and brag worthy....
     
  11. conroe

    conroe Member

    Joined:
    Apr 19, 2007
    Messages:
    341
    Location:
    Sydney
    ....... double post
     
  12. conroe

    conroe Member

    Joined:
    Apr 19, 2007
    Messages:
    341
    Location:
    Sydney
    Well if you know super pi does not prove anything about the cpu power then why the *@%k carrying on like an idiot. What a hypocrite.

    If you want to know how good the AMD Hex is run LinX. If you can get even close to 50 GFlops with your 850 I'll shut the hell up.

    Oh... just in case you're wondering, my conservatively oc 1055T achieves 67 GFlops and I can assure you you'd be lucky to get 45.
     
    Last edited: Jul 7, 2010
  13. mevereyn

    mevereyn Member

    Joined:
    Dec 29, 2009
    Messages:
    1,010
    ffs.

    (10C)
     
  14. BuuBox

    BuuBox Member

    Joined:
    May 7, 2003
    Messages:
    9,292
    Location:
    Melbourne
    Using Y-Cruncher my Q6600 does 1M in 0.699 seconds.

    :lol:
     
  15. Psychotik

    Psychotik Member

    Joined:
    Jan 2, 2005
    Messages:
    2,472
    Location:
    127.0.0.1
    Exactly. My 4 cores at a very conservative OC beats AMD's 6 cores at a pretty high overclock!
     
  16. OP
    OP
    m1one

    m1one Member

    Joined:
    Oct 7, 2009
    Messages:
    197
    Hmmm

    I thought this was a AMD thread ?

    If I was after a CPU to do Pi bragging , yes I would buy a Intel .
    I posted this result as I would be comparing it to other AMD cpu's , its simply a given that intel chips eat AMD @ Pi .

    Unfortunately Hwbot.org does not run anything favoring AMD .
     
  17. Archy

    Archy New Member

    Joined:
    Jan 23, 2010
    Messages:
    1,312
    Location:
    Glenhaven, Sydney
    lol, pun fixed.

    It's a shame, yeah. And I'm assuming it's well known because it's so old and a newer, more efficient program will calculate 1M way too quickly. If you are dealing with >1 sec times, I imagine that a margin of error would make a much bigger difference.
     
  18. SingAlong

    SingAlong (Banned or Deleted)

    Joined:
    May 30, 2010
    Messages:
    474
    Location:
    Sydney 2035
    At a bigger overall price... have you seen the price of the 1055T?!?!?

    That benchmark was maybe relevant in the days of the 300A@450, but as mentioned a few times throughout the thread, is mostly used for E-Peen nowadays :lol:

    I actually thought this was the AMD Forums? :eek: :lol:
     
  19. mAJORD

    mAJORD Member

    Joined:
    Jun 4, 2002
    Messages:
    9,831
    Location:
    Griffin , Brisbane
    <Translated> "I'm going to keep trolling, despite being out of ammo"



    :p
     
  20. Copie

    Copie Member

    Joined:
    Jul 24, 2009
    Messages:
    13,656
    Location:
    Newcastle
    Cool story, if i actually stopped my DC++ hashing + watching star trek HD i would probably get better as well.
     

Share This Page

Advertisement: