Results 1 to 9 of 9

Thread: 28433 sieve

  1. #1

    28433 sieve

    Hi All

    The latest result for k=28433 sieve up to 318G by Joseph Arthur is attached.

    If Louie have not started his sieve on this special k-value it is free for everyone to start the sieve.

    Joseph have reserved the k=28433 up to 2,000,000 and currently he had done all the work up to 1,810,000. you can see his progress at
    http://robin.mathi.com/28433

    Regards
    Payam
    Attached Files Attached Files

  2. #2
    If nothing speeks against it i will do sieving up to 400G

    Lars

  3. #3
    i'd recommend against sieveing this k. as it is, i have already sieved beyond the limit of the posted file. I downloaded it awhile ago directly from Arther's site. I am close to 400G already. thanks anyway.

    looking things over, if you want to sieve something, I think k=55459 for n < 3000000 between 200G and 300G would probably be a better use of resources... I'm not sure exactly how that one didn't get sieved a little higher but that would be the best place to do work right now.

    -Louie
    Last edited by jjjjL; 12-25-2002 at 11:59 AM.

  4. #4

    Help Sieving

    I've been trying to find a SB client for an UltraSparc II. I don't think it exists (yet?). Barring that I have 4 4xUS2 boxes with nothing to do. Can these sieve with current binaries?

    I'm quite new to this, so any help would be good!

    Thanks

  5. #5
    Lucky enough i had not really started sieving k=28433.

    So i will stay with k=4847.

    Lars

    Edit:

    @Louie:

    Two Questions about K=55459.

    First question Is only the range from 200G to 300G needed ?
    And second question where could one get the needed infile?

    If i get the infile i would like to go 200G to 300G.


    Last edited by ltd; 12-25-2002 at 03:37 AM.

  6. #6

    Re: Help Sieving

    Originally posted by Cowering
    I've been trying to find a SB client for an UltraSparc II. I don't think it exists (yet?). Barring that I have 4 4xUS2 boxes with nothing to do. Can these sieve with current binaries?

    I'm quite new to this, so any help would be good!

    Thanks
    I was working on a port, but with the recent rate of prime discoveries (about one a week!), I no longer consider it a wise investment of time (at the moment).

    The current SoB client has an x86 assembler core. The current non-x86 version runs about 30 times slower, and is using a c math library.

    If you have coding skills, and have the time, contact Louie.

    Otherwise, for the UltraSparc II boxen, I suggest running distributed folding as Dyyryath has a very nice stats page.

  7. #7
    I was working on a port, but with the recent rate of prime discoveries (about one a week!), I no longer consider it a wise investment of time (at the moment).
    Don't get fooled by the rate we find the primes. There are still 12 more primes to be found. We only found the 'easy' ones.

    To test a number that is twice as large, takes about 4 times longer to test.

    But a client for non x86 platforms must have a reasonable speed. Something like within 50% of the SoB client

  8. #8
    Member
    Join Date
    Dec 2002
    Location
    Eugene, Oregon
    Posts
    79

    long-term sieving strategy

    It sounds like Louie has sieved most of these candidates to around 400G or a little more. On the other hand, it looks like Sanders and friends have sieved k=4847 from 200G to 1T and so have eliminated about 5% of the candidates between 2M and 3M. That suggests to me that the ideal sieving limit for all these k values should be several times larger than current limits, maybe even 5T to 10T. Unfortunately, we are getting close to starting on the n ranges from 2M to 3M soon, so there won't be time to sieve that far. However, we also need to be ready with well-sieved values when we reach 3M. I think that it does not make sense to not take advantage of the power of distributed computing for the sieving part of the project. However, the files involved will be large, maybe 1Mbyte or so, and it would probably make the most sense for Louie to use Paul's new multiple k sieve on a range that would keep us busy for a couple of years, maybe 3M to 20M (or even higher). Other people who want to sieve could work further in the meantime on the 2M to 3M range until we start making headway past 2M.

    Most candidates get removed in the early stages of sieving, which is why it would be better for that work to be done by Louie on the University cluster. Once he gets as far as 200G, my guess is that he will still have around 700,000 candidates left with n between 3M and 20M. Such a master file can be compressed to around 700 Kbytes, and in fact, sievers won't have to update their working master file that often, since they will just report back with lists of the values they have eliminated. But given that around 20M, we want to be sieving way, way beyond our present sieving limits, maybe even above 500T, it seems to me that the total ongoing sieving effort needs to be at least a few per cent of the total project effort. I also see this as an advantage to the project in that some people are attracted to a project where they see some division of labor in the total effort. (For example, a number of people in GIMPS specialize in doing only factoring.)

  9. #9
    I made some tests with k=4847 and n between 3M and 20M.
    After sieving only up to 8M there are only 115000 numbers left and numbers are still removed wirth a rate of 16n per second.

    So the files which have to be distributed wouldnd be to big.
    ( At the moment 1.5MByte uncompressed)

    Lars

Posting Permissions

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