Loading ...
Sorry, an error occurred while loading the content.

Help please... user or pgm bug with NewPGen 2.82 on AMD & P4

Expand Messages
  • lasher_chuck
    This is probably THE most embarassing question.... but I am perplexed to the max. May I (a novice at this) ask your indulgence for some educating /
    Message 1 of 4 , Jan 30, 2005
    • 0 Attachment
      This is probably THE most embarassing question.... but I am perplexed
      to the max.


      May I (a novice at this) ask your indulgence for some educating /
      verification of a condition?


      I have two machines that do most work...

      AMD-64 FX series for sieving (16 MP/sec not uncommon).

      P4b 2.4G for LLR (ver 3.3 currently).



      I have been trying to learn / catch up.

      I sieved, for 24 hours as is (I am told) the norm using NPG.

      I ran the NPG file and got nothing. Sieved the next block, got
      nothing. I went back to the beginning (low N) and started again.

      Again... there were no primes found.


      After the 4-5th go around, I got deeper into this and went in small
      increments...

      I did the most rudimentary test: k.b^n-1
      using: k=7, b=2, 2<=n<=65536

      @ P=1G, K=7,N=29 was in the NPG as the lowest file.
      @ P=5G, it was gone... as were MANY of valid K N
      (which would yield prime) values.


      This has been verified on seperate processors (Intel and AMD).
      All machines are Prime95 (Gimps) full torture test certified.

      They were also recertified prior to the start of this test run.



      Am I using the wrong software; sieving too much; or a known issue?

      The Engineer in me says that there smells of a 2^32 wrap around bug @
      the P=4G mark w/ or w/o the combo of 2 & 65536 for N.



      PS: Since I *SHOULD* know better and be able to solve this myself,
      please don't hestitate to slap me around either... LOL :)


      I thank you for your time. I thank you in advance for your
      assistance.



      Chuck
    • jim_fougeron
      ... I am pretty sure that what you are seeing, is that NPG has factored the number. The factor found, was the number itself. NPG has made this choice to NOT
      Message 2 of 4 , Jan 30, 2005
      • 0 Attachment
        > I did the most rudimentary test: k.b^n-1
        > using: k=7, b=2, 2<=n<=65536
        >
        > @ P=1G, K=7,N=29 was in the NPG as the lowest file.
        > @ P=5G, it was gone... as were MANY of valid K N
        > (which would yield prime) values.
        >

        I am pretty sure that what you are seeing, is that NPG has factored
        the number. The factor found, was the number itself. NPG has made
        this choice to NOT check to see if a found factor is "itself", due
        to checks like that would slow down the overall speed of NPG, and
        it would be assumed that very tiny N such as this would be proven
        prime/composite using other trivial methods.

        It is simply a run time behavior of NPG which you learn to live with.
        If you were concerned about getting those tiny numbers, then I would
        recommend using PFGW, and create this file:

        ABC2 7*2^$a-1
        a: from 1 to 1000

        (call the file pfgw.in) Then perform a pfgw -f pfgw.in. There will
        be 2 files, a pfgw-prime.log (proven primes proven by trial division)
        and a file pfgw.log with the PRP-3's. Then simply run
        pfgw -tp pfgw.log and all the primes will be in pfgw-prime.log.

        Then use NPG from n=1000 up to whatever limit you want.

        Or you could simply look on the web page:
        http://www.prothsearch.net/riesel2.html
        and find all known 7*2^n-1 up to n=1100000

        Jim.
      • Jens Kruse Andersen
        ... The norm for any sieve is to sieve until the removal rate is similar to the prp/test time (which may be a complicated issue for a fixed k sieve). ... The
        Message 3 of 4 , Jan 30, 2005
        • 0 Attachment
          Chuck wrote:

          > I sieved, for 24 hours as is (I am told) the norm using NPG.

          The norm for any sieve is to sieve until the removal rate is similar to the
          prp/test time (which may be a complicated issue for a fixed k sieve).

          > I did the most rudimentary test: k.b^n-1
          > using: k=7, b=2, 2<=n<=65536
          >
          > @ P=1G, K=7,N=29 was in the NPG as the lowest file.
          > @ P=5G, it was gone... as were MANY of valid K N
          > (which would yield prime) values.

          The prime 7*2^29-1 = 3758096383 is between 1G and 5G (G = 10^9).
          As Jim also thinks, NewPGen probably removed it when the prime was found as
          factor of itself.
          Some of my sieves knowingly do the same.
          Users and programmers can argue about whether something is a bug or feature.

          If NewPGen also removes some known primes above the sieve limit:
          Can you give examples?
          Have you checked "Verify results" ?

          The help says:
          If "Verify results" is checked, NewPGen will do a careful check to make sure
          that the prime actually does divide the number that it thinks it does. This
          will make NewPGen very slightly slower, but it is recommended that it is
          always checked, particularly as p gets large (where the scope for programming
          errors increases)

          If you check "Log the numbers removed" in the Options menu then NewPGen will
          write which prime removes a candidate, but only for primes above 2^32.

          --
          Jens Kruse Andersen
        • Paul Jobling
          ... As other have stated, those numbers are being removed because the numbers themselves are being found as factors. I could change the software to catch that;
          Message 4 of 4 , Jan 31, 2005
          • 0 Attachment
            > I did the most rudimentary test: k.b^n-1
            > using: k=7, b=2, 2<=n<=65536
            >
            > @ P=1G, K=7,N=29 was in the NPG as the lowest file.
            > @ P=5G, it was gone... as were MANY of valid K N
            > (which would yield prime) values.

            As other have stated, those numbers are being removed because the numbers
            themselves are being found as factors. I could change the software to catch
            that; thinking about it it might be fairly easy to perform one additional
            check while removing from the bitmap.

            Regards,

            Paul.


            __________________________________________________
            Virus checked by MessageLabs Virus Control Centre.
          Your message has been successfully submitted and would be delivered to recipients shortly.