Page 3 of 3 FirstFirst 123
Results 81 to 116 of 116

Thread: Sieve Coordination Thread [old]

  1. #81
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    Ah, ok, thx!
    Then I'll take 275-280

    So we should have now:

    0 - 25G Louie
    25 - 50 ceselb
    50 - 75 kmd
    75 - 100 cjohnsto
    100 - 125 Nuri
    125 - 130 McBryce
    130 - 150 RangerX
    150 - 155 nuutti
    155 - 175 MikeH
    175 - 200 paul.jobling
    200 - 210 dudlio
    210 - 215 smh
    215 - 225 dmbrubac
    225 - 250 EggmanEEA
    250 - 255 frmky
    255 - 260 Alien88
    265 - 275 ltd
    275 - 280 Mystwalker

    300-301 Paperboy


    Already any measures of what are good alpha factors?

  2. #82
    Moderator ceselb's Avatar
    Join Date
    Jun 2002
    Location
    Linkoping, Sweden
    Posts
    224
    Originally posted by Mystwalker
    Already any measures of what are good alpha factors?
    The defults are ok, I think. This varies over the ranges, so a quick test might be in place. Shouldn't be too far off though, so do as you want.

  3. #83
    Junior Member
    Join Date
    Jan 2003
    Location
    Netherlands
    Posts
    2
    I was playing around with the programs, and it seems that Nbegon64 is more than twice as fast on my xp 2000+. I've done some testing on the range 301 - 302 range, so I'll take 301 - 310 for now.
    That leaves a gap between 280 and 300.

    With fatphil's client I'm currently using an alpha of 1.000. This setting was the fastest for me.

  4. #84
    Originally posted by MikeH
    Phil,

    Just to check the .bat save file will work as a service, I just tried the following line in a .bat file

    nbegon64 -p170886000000-175000000000 -ssob.dat

    Launching the .bat file with FireDaemon is a breeze. I look forward to tomorrow's version.
    Out at a gig tonight ("The Crown" from Sweden - amazing!) so I won't be up till /late/ tomorrow...

    I decided that a parameter file was a good idea, and I also decided that the best format for a parameter file was to look like a batch file command line! i.e. I won't need to change the code I've already written. I'm glad the batch file works as is though.

    Bed...
    Phil

  5. #85
    Hey I answered my own question about the files. I don't know if the info will be useful to anyone else but in XP at least the file writing is saved as it writes them. So in other words if you're running the program and say... you're computer crashes/freezes, then all you have to do to start over is open up the current SoB.del file and start from the last p listed there.

    That's why I was asking, because I'd prefer to run the program on the full range instead of breaking it up to make sure I don't lose anything in the case of a freeze. But now I don't have to worry about it.

    PS. Of course the addition of the batch file generator makes all of this completely useless

  6. #86

    OK, point taken, SoBSieve improved

    Well done Phil on the NBeGone stuff, it really pushed me into looking at SoBSieve and the bottlenecks in there. I found one, and now I think that - while I don't have access to every platform - this new version of SoBSieve should do the business.

    NOTE that alpha should be set to 1 now!

    This also automatically starts sieving when you set it off. I will be looking to get more performance out of this as well, as NBeGone pushes it close for *very* large values of p... Phil, how about I take Windows and you have the rest of the World?

    Regards,

    Paul.
    Attached Files Attached Files

  7. #87
    Nice speed up in the sobsieve client. It seems to be 2x or more as fast as the 1.07 client on my p3

  8. #88
    On my Athlon XP 1800+, the new SOBSieve runs at 26450 p/s which is MUCH better than the previous 11890 p/s and a bit faster than NbeGon's 24485 p/s. Good work!

    Greg

  9. #89
    Paul does it again! I knew he could.

    My Athlon 1.33GHz went from 9500p/sec -> 19k p/sec.

    I updated the download on the sieve page:

    http://www.seventeenorbust.com/sieve/

    The new download has a slightly higher sieve file as well. There is also a new link to Phil's site.

    DON'T FORGET: set alpha back to 1 for the new SoBSieve.

    -Louie

  10. #90
    Thanks Paul, with Alpha=1 SoBsieve works about 4% faster on my PIII450 compared to NbeGon64.

    I had some problems starting the client though. Because i was running NbeGon64 i renamed the status file. After starting SoBSieve110 i got a popup to enter the range to test, but after clicking OK, the popup won't go away and the sieving won't start.

    I created a new status file with an old version of SoBsieve and after that starting the newest version it gives a popup with the range i entered in the old client, i press ok and the client starts sieving.

    So, to me it looks like the newest version needs a statusfile, but doesn't create one itself.

  11. #91

    Re: OK, point taken, SoBSieve improved

    Originally posted by paul.jobling
    Well done Phil on the NBeGone stuff, it really pushed me into looking at SoBSieve and the bottlenecks in there. I found one, and now I think that - while I don't have access to every platform - this new version of SoBSieve should do the business.

    NOTE that alpha should be set to 1 now!

    This also automatically starts sieving when you set it off. I will be looking to get more performance out of this as well, as NBeGone pushes it close for *very* large values of p... Phil, how about I take Windows and you have the rest of the World?
    Take windows. Where are you going to take it.

    However, purely for completeness I compiled a windows version of my latest version, just in case anyone wants a copy.

    Available from
    http://fatphil.org/maths/sierpinski/bin/
    are the following
    007 - with the resume data (put into a file 'SoB.bat')
    008 - like 007, but faster too! (linux/win/sun/alpha so far)

    It would be silly for me to give a figure comparing speed ratios, given how weird my machine behaves on SoBSieve, but shall we just say that I wouldn't be running 007 if I were you!

    I have only tested using my old favourite ranges (I can verify them by inspection now as I've seen them so many times), but it would be nice if someone who did one of the lower ranges (more factors) could either verify, or issue a bug report...

    Note - the optimal dimensions have changed: -d=2.9 works well for me, but 2.7-3.1 were pretty flat. It'll depend on architecture

    Phil

  12. #92


    Why don't you two exchange code to make one SUPER fast siever!


    Btw do you guys profile your executables? I heard that can boost performance considerably too.

    Cheers,

    Ola

  13. #93
    Phil,

    There is an obvious need for a non-Windows sieving client and your software fills that role very well. But for Windows I have got many optimisations to come and it is crazy to be this competitive - I have got an easy 10% improvement with a couple of one line changes, but I want to spend some time getting it all sorted and solid and get further improvements rather than rushing out updates in an optimisation battle.
    Cheers,

    Paul.

  14. #94
    Why don't you two exchange code to make one SUPER fast siever!
    I offered to do exactly that, but Phil didn't want to play ball . Which just spurred me on...

    Cheers,

    Paul.

  15. #95
    Great, back to Phil's N-be-Gone

    with -d=1,3 (works faster then -d=3 for me) it's about 40% faster compared to SoBsieve110

    It's over 3 times faster compared to the original SoBsieve

  16. #96
    Moderator ceselb's Avatar
    Join Date
    Jun 2002
    Location
    Linkoping, Sweden
    Posts
    224
    My main reson for switching to NbeGon was the long time to complete my range. With these new versions it's now down to managable times (32 vs. 15 days). Now I've switched back to SoBSive again.

    Advantages of the two programs:
    SoBSieve
    • GUI.
    • No need to tinker with command lines or shortcuts / scripts.
    • Rate is viewable.


    NbeGon
    • Run as service / at startup.
    • No need to remove lines from output file.
    • Runs on non-windows boxes.

  17. #97
    Hi,

    only to correct the last list of reserved ranges.
    I have reserved 260 to 275.

    Lars

  18. #98
    I've got a version in progress that already has a 40% improvement over the last version- and that is at very high values of p where the software was (relatively) weak before. I have also fixed the problem with having to click four times when it in in the system tray. I'll keep on improving this before sending out a 1.11 release.

  19. #99
    Originally posted by paul.jobling
    I offered to do exactly that, but Phil didn't want to play ball . Which just spurred me on...
    I seem to remember you inquiring if we could merge my hashing with your maths, and I said that I can't split my code easily, it's too tightly coupled (do one job, and do it well). I did say I could offer you a doOneP(long long p) interface, and if that wouldn't do I asked exactly where you would like me to try and cleave it, for me to mull over while I'm in the code, but you didn't reply.

    The _only_ cleaving I can currently do is to separate the filling of the hash table from the rest of the code. My 52+bit maths is very clumsy (as it's C), and I'm sure that you could speed that up noticably.

    My current test version is much faster than 008, and even then I've not run out of ideas, and the code's getting more chaotic as I try these new things out - i.e. my algorithm is still changing. However, what I'm also doing while going through the code is stripping out the dead wood, the things that failed. So hopefully I will soon end up with a version that contains just what is needed and nothing more, and that's possibly something that you can work with. However, as I say I see nowhere to merge algorithms, as mine's pretty much one monolithic block.

    Mail me, we can talk algorithms. If I can do it in C you can do it faster in assembler, but I genuinely believe it makes sense to get the algorithm correct before forking too much of it onto an assmebly language route.

    Phil

  20. #100
    The reason i installed the fastest version available at this moment is that i'm going on a vacation this weekend and i want to keep the program running on my office computer.

    If any faster version will be available by tomorrow, i'll install that one, since i won't be able to upgrade for the next 4 weeks (although i doubt my pc will keep on running for that long).

  21. #101
    Originally posted by smh
    The reason i installed the fastest version available at this moment is that i'm going on a vacation this weekend and i want to keep the program running on my office computer.

    If any faster version will be available by tomorrow, i'll install that one, since i won't be able to upgrade for the next 4 weeks (although i doubt my pc will keep on running for that long).
    I'll have my fastest algorithm (i.e. no plans for any new major changes) ready by this evening. After that it I'll have nothing but micro-optimisations to do (loop unrolling etc. that might be platform specific) but it makes sense to draw a line under it before I contort the C out of current relatively simple state.

    Phil

  22. #102
    Senior Member dmbrubac's Avatar
    Join Date
    Dec 2002
    Location
    Ontario Canada
    Posts
    112

    wonderful work phil and paul

    My completion estimate has dropped from 33 days to 12 due to program optimizations! I tried Phil's for a while but I'm back to Paul's (1.10) because sobseive shares it toys better with sb.

    Thanks again guys

  23. #103
    Sieve it, baby!
    Join Date
    Nov 2002
    Location
    Potsdam, Germany
    Posts
    959
    @ltd:

    Oops, bad typo. I'm sorry!

    So here's the corrected table:

    0 - 25G Louie
    25 - 50 ceselb
    50 - 75 kmd
    75 - 100 cjohnsto
    100 - 125 Nuri
    125 - 130 McBryce
    130 - 150 RangerX
    150 - 155 nuutti
    155 - 175 MikeH
    175 - 200 paul.jobling
    200 - 210 dudlio
    210 - 215 smh
    215 - 225 dmbrubac
    225 - 250 EggmanEEA
    250 - 255 frmky
    255 - 260 Alien88
    260 - 275 ltd
    275 - 280 Mystwalker

    300-301 Paperboy

    btw. I'm using NbeGon_008 with an d=0.8 as it's the fastest setting for me so far.
    There are ~15x as many overflows as with d=2.8, but only in the first column.

  24. #104

    Re: wonderful work phil and paul

    Originally posted by dmbrubac
    My completion estimate has dropped from 33 days to 12 due to program optimizations! I tried Phil's for a while but I'm back to Paul's (1.10) because sobseive shares it toys better with sb.

    Thanks again guys
    Have you tried setting the priority to "idle"?

    Phil

  25. #105
    Junior Member
    Join Date
    Jan 2003
    Location
    Netherlands
    Posts
    2
    I'm currently working on 301 - 302, as I posted yesterday. In that post I decided to take 301 - 310. Is is better to stop after the program has finished the 301.x range and then take 280 - 290, or should I go on with 302 - 310?

  26. #106
    I've got one, possibly two, machines without internet connections that could be used for sieveing. So i know where to get SoBseive or Nbegone... and where to post results... so where/who do i get SoB.dat

  27. #107
    Senior Member dmbrubac's Avatar
    Join Date
    Dec 2002
    Location
    Ontario Canada
    Posts
    112
    Have you tried setting the priority to "idle"?
    Yes I did. I suspect that since there are actually 31 levels of relative thread priority under windows that there is a minor difference somewhere.

    For instance, if both are running at "idle" Process Priority Class (which is, I believe, basically what you are changing when you modify the priority in Task Manager and exactly what you are changing when you modify the priority in SOBSeive), the threads could still be at different Thread Priorities (Time Critical, Highest, Above Normal, Normal, Below Normal, Lowest, Idle).

    I may not be exactly correct here, but I think I'm close. Look at Analyzing Processor Activity for more info.

    HTH

    Dave

  28. #108
    Senior Member dmbrubac's Avatar
    Join Date
    Dec 2002
    Location
    Ontario Canada
    Posts
    112
    so where/who do i get SoB.dat
    Download SOBSeive. Even if you don't use it, it will come with the SoB.dat file

  29. #109
    I'm done sieving 255-260.

  30. #110
    i'll take 285-290

  31. #111
    This is getting a bit confusing, so I repost the whole thing again:

    0 - 25 Louie
    25 - 50 ceselb
    50 - 75 kmd
    75 - 100 cjohnsto
    100 - 125 Nuri
    125 - 130 McBryce
    130 - 150 RangerX
    150 - 155 nuutti
    155 - 175 MikeH
    175 - 200 paul.jobling
    200 - 210 dudlio
    210 - 215 smh
    215 - 225 dmbrubac
    225 - 250 EggmanEEA
    250 - 255 frmky
    255 - 260 Alien88
    260 - 275 ltd
    275 - 280 Mystwalker

    285 - 290 alexr

    300 - 301 Paperboy
    301 - 310 Zonar


    --> Two people needed for the remaining gaps!! (I cannot do it,
    my processor is way too slow)

  32. #112
    I'm done with 300-301

  33. #113
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    All this competition does seem to be good! Good work Paul, good work Phil.

    SoBSieve 1.1 running on an AMD XP 2100+ = ~40Kp/sec

    Paul, what should I now be using for the alpha? Currently I'm using 1.0. Or should I just experiment?

    One more request - we are half way there will being able to put SoBSieve in Windows start-up group, but we still get the dialogue box that asks "The sieving is from p = ... to p = ... - OK to go?", and you have to click yes. Please can you remove this dialogue box, and maybe display the range in the title bar "SoBSieve 1.1 (170.2G - 175.0G)", or maybe just display this somewhere else on window once it's running. But key point is - it should start without user interaction.

    And can you clarify what happens on power failure? Does it pick up the last pmin= in the SobStatus file, or does it take the last factor in the file and pick up from there? Or something else?

  34. #114
    Senior Member
    Join Date
    Jan 2003
    Location
    UK
    Posts
    479
    Has anyone else tried to get Phil's NbeGon_008 (or the batch file it produces to be more precise) running as a service with FireDaemon?

    It works fine with Win2K, but under WinNT using the same setting it doesn't work. Symptom is FireDaemon says the service has started, but when I look in task manager it isn't! Any clues anyone?

  35. #115
    I'm grabbing 280-285 now..


    0 - 25 Louie
    25 - 50 ceselb
    50 - 75 kmd
    75 - 100 cjohnsto
    100 - 125 Nuri
    125 - 130 McBryce
    130 - 150 RangerX
    150 - 155 nuutti
    155 - 175 MikeH
    175 - 200 paul.jobling
    200 - 210 dudlio
    210 - 215 smh
    215 - 225 dmbrubac
    225 - 250 EggmanEEA
    250 - 255 frmky
    255 - 260 Alien88 [complete]
    260 - 275 ltd
    275 - 280 Mystwalker
    280 - 285 Alien88
    285 - 290 alexr


    300 - 301 Paperboy [complete]
    301 - 310 Zonar

  36. #116
    I would like to divide this discussion into two new threads, so I am going to lock this one and start a new one.

    The new ones will be:
    Sieve Coordination Thread - please use this exclusively for coordinating your blocks
    Sieve Client Thread - please use this to discuss the clients.

    Thanks,
    Alien88

Page 3 of 3 FirstFirst 123

Posting Permissions

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