P8101 is now mainstream -bigadv

Discussion in 'Distributed Computing' started by Nathan_P, Apr 22, 2012.

Thread Status:
Not open for further replies.
  1. MrValentine

    MrValentine [H]ard|Gawd

    Messages:
    1,031
    Joined:
    Jul 5, 2004
    If that is true, then a REVERSE core hack could get you more PPD on a 4p box. Trick the core into thinking you have 12 threads instead of 32 and you get a 6903.

    If that happens i'm going to fall over laughing as thats embarrising for Stanford.
     
  2. Zink

    Zink Gawd

    Messages:
    693
    Joined:
    Feb 9, 2010
    The reverse core hack would give you the same performance as 12 threads though. You could get the 12 thread units but with lower performance. Running multiple 6903 might come close to a single 8101 but that's not what Stanford wants so there isn't really any reason to try it out.
     
    Last edited: Apr 23, 2012
  3. Linden

    Linden [H]ard|Gawd

    Messages:
    1,140
    Joined:
    Sep 8, 2005
    Zink, I don't understand you. The "core hack" (script) only provides system information to the Stanford assignment servers. It does not limit CPU engagement. It neither increases nor decreased a system's performance on any given work unit.

    I advise neither for nor against using the script.
     
  4. 402blownstroker

    402blownstroker [H]ard|DCer of the Month - Nov. 2012

    Messages:
    3,156
    Joined:
    Jan 5, 2006
    After poking around a little on the FF site, I am guessing the 8101 WU is setup as a punishment for doing the corehack trick. Also looks like Stanford/PandeGroup does not care if you did it or not, but everyone will suffer because it was used.
     
  5. Nathan_P

    Nathan_P [H]ard DCOTM x2

    Messages:
    3,077
    Joined:
    Mar 2, 2010
    I have to agree with this, if we didn't have 2500k/2600k and x6's doing -bigadv we would be fine.
     
  6. Biffa

    Biffa Limp Gawd

    Messages:
    226
    Joined:
    Jan 7, 2011
    Pulling 00:13:54 333366 PPD on my quad 6176
    Pulling 00:25:39 133367 PPD on my X5650 SR-2#

    For comparison, AMD rig gets near 500K on a 6903 and the SR-2 around 160K
     
  7. Vaulter98c

    Vaulter98c [H]ard|DCer of the Month - October 2009

    Messages:
    5,097
    Joined:
    May 21, 2008
    so it's not such a hit to lower end rigs like 2P's, but on 4P's, especially fast ones, it's a swift kick in the balls? Kinda defeats the whole point of the QRB right lol
     
  8. Biffa

    Biffa Limp Gawd

    Messages:
    226
    Joined:
    Jan 7, 2011
    33% loss on the 48 core and 20% loss on the 24 thread :rolleyes: :(
     
  9. Untitledone

    Untitledone [H]ard|DCer of the Month - April 2012

    Messages:
    1,500
    Joined:
    Feb 28, 2011
    We feel you biffa. For about the same amout of time, a 6903 gets me 514K PPD, and an 8101 gets me 350K PPD, 32% reduction. (13:30 TPF 0.94 days completion 8101 vs 13:51 0.96 days completion 6903)
     
  10. Linden

    Linden [H]ard|Gawd

    Messages:
    1,140
    Joined:
    Sep 8, 2005
    More data.

    G34 4P, 32-core, 4 X 6136, 12.5% OC, 2.4 @ 2.7GHz:
    Typical TPF & PPD for usual projects 6901 and 6903, plus first instance of 8101 --
    6901, 07;45, 274,000
    6903, 17:11, 372,200
    8101, 16:42, 253,900

    So, at least with this system, if compared to 6903 processing, which requires approximately the same system resources, as measured by time/frame, points yield is 32% less. Points yield, as compared to a 6901 (a 12-core qualifying unit), is 7% less.

    If the intent for this project was to be difficulty/award parity with the original bigadv project 6901, they aren't too far off. (Is 6900 considered bigadv?)

    I really don't know what my opinion is on the change. I was one of the users, who a few months back, was very concerned about the ever increasing PPD inflation. On the other hand, my emotional reaction now is, what on Earth...!?

    At present, I am not prepared to render a judgment of 'good' or 'bad,'
     
  11. Grandpa_01

    Grandpa_01 [H]ard|DCer of the Year 2013

    Messages:
    1,157
    Joined:
    Jun 4, 2011
    Actually the 8101 is the result of a few people repeatedly griping about the QRB and starting threads about the unfairness of the point system. There were a few of us mostly Nathan P, Patriot and myself that have tried to counter there post over the last year or so, but unfortunately the other side was able to convince Stanford that the bigadv folders were getting too many points and that the points system was out of whack and needed to be adjusted.

    They used all kinds of different arguments to get there point across and I failed at convincing Stanford otherwise we really did not have allot of support from the folding community itself, I believe this is due to the fact that most of the folders never visit the FF anyway, and the proponents were asking folders from there perspective forums to come and post on the FF in the threads about the point system.

    In hind site I probably should have done the same and who knows perhaps we would not be where we are now.
    :(
     
  12. Zink

    Zink Gawd

    Messages:
    693
    Joined:
    Feb 9, 2010
    This could be wrong but from googling: The "core hack" actually provides information to the client about how many simultaneous threads the CPU can handle. In google images I can see people who have the hack active on a 2500k show that the WU maps to 8 threads that all run at once on the 4 cores. If you tell the client you only have 12 or 15 cores then it will only run that many threads and you will need multiple clients to saturate the CPUs. It's probably a bad idea but if PPD was slightly higher that would make Standford look stupid.
     
  13. Grandpa_01

    Grandpa_01 [H]ard|DCer of the Year 2013

    Messages:
    1,157
    Joined:
    Jun 4, 2011
    No the fah client will use it to use as many cores as you tell it to use if you set it to -smp 48 it will use 48 cores. The core hack has nothing to do with the fah client or folding it just reports the core count as what ever you tell it to report.
     
  14. extide

    extide 2[H]4U

    Messages:
    3,155
    Joined:
    Dec 19, 2008
    You dont need any hack to report less cores. On a 48 core box you could run 4 instances of -smp 12 if you wanted.
     
  15. Linden

    Linden [H]ard|Gawd

    Messages:
    1,140
    Joined:
    Sep 8, 2005
    Grandpa, I was one of the people, who about 8 or 9 months ago, expressed my concern about the rapid rate of increase of points awards. You probably read my comments at FF, which I hardly believe were "griping." (you're still my friend ;)) Low and behold, Pande Group did adjust the QRB K factor a while later, which, at the time, I thought was a reasonable action. Perhaps I wasn't paying close enough attention, but I thought the first adjustment was the fix, so to speak. Except for the beta testers, I don't think any of us saw this next (8101) adjustment coming.
     
  16. Kendrak

    Kendrak [H]ard|DCer of the Year 2009

    Messages:
    22,871
    Joined:
    Aug 29, 2001
    It has been talked about in the DAB (and is still)

    Might I suggest letting Stanford know what you all think of the 8101 by starting a thread in the FF. (They don't read too much in the team forums, and I can't blame them, there are quite a few.)

    Please be respectful and kind. State your case and stats to back it up. Pitchforks and torches don't help.

    I am trying in the DAB. I am it seems outnumbered.
     
  17. Core32

    Core32 [H]ard|Gawd

    Messages:
    1,053
    Joined:
    Mar 3, 2012
    Can you provide a link to that particular discussion?
    Thanks.
     
  18. Kendrak

    Kendrak [H]ard|DCer of the Year 2009

    Messages:
    22,871
    Joined:
    Aug 29, 2001
    I delved into FF (shudder) and found this thread http://foldingforum.org/viewtopic.php?f=16&t=21036

    I think this is what they are talking about.

    I'm going back into Genmay where it is safe. :eek:
     
  19. tjmagneto

    tjmagneto [H]ard DCOTM x2

    Messages:
    3,007
    Joined:
    Aug 6, 2008
    He just shrugged the matter off and went about business as usual.
     
  20. Skillz

    Skillz [H]ard DCOTM Jan,2016

    Messages:
    15,850
    Joined:
    Aug 14, 2004
    Well so far we're still putting up more PPD than EVGA. When is this supposed to start effecting us?
     
  21. Kendrak

    Kendrak [H]ard|DCer of the Year 2009

    Messages:
    22,871
    Joined:
    Aug 29, 2001
    They are in the wild now.
     
  22. Untitledone

    Untitledone [H]ard|DCer of the Month - April 2012

    Messages:
    1,500
    Joined:
    Feb 28, 2011
    The day [H]ardocp gives up, takes it in the ass, and loses unity and integrity.
     
  23. Skillz

    Skillz [H]ard DCOTM Jan,2016

    Messages:
    15,850
    Joined:
    Aug 14, 2004
    I meant when should we start seeing points fall, as long as we still produce more PPD than EVGA we will eventually catch them. So far, we're still doing more PPD than they are.
     
  24. Zink

    Zink Gawd

    Messages:
    693
    Joined:
    Feb 9, 2010
    I just got a 6904 which I haven't seen since last month.
     
  25. Untitledone

    Untitledone [H]ard|DCer of the Month - April 2012

    Messages:
    1,500
    Joined:
    Feb 28, 2011
    I knew what you meant, I just wanted to throw in some inspiration :). EVGA does have -bigadv machines, so they will be effected as well. We will have to wait and see how we are all effected by this before we come to any conclusions.
     
  26. Grandpa_01

    Grandpa_01 [H]ard|DCer of the Year 2013

    Messages:
    1,157
    Joined:
    Jun 4, 2011
    Probably will not notice it due to recent additions to the team Atlas / tear is back on line it looks like core32 may have added a little something ChelseaOilmans recent addition of some faster chips and that is just the ones I know of. Some of the MP rigs or all will be running the core hack or other work around. It is probably enough to maintain our daily production at least for a while. or maybe even increase it. But it is mainly due to increased cpu cycles. If all of these thing were not happening then the production would probably decrease somewhere around 10% to 15%
     
  27. Skillz

    Skillz [H]ard DCOTM Jan,2016

    Messages:
    15,850
    Joined:
    Aug 14, 2004
    Well if I didn't have so many projects going at once, I would have built a 4p by now. (or at least have the money waiting to purchase parts)
     
  28. Patriot

    Patriot [H]ard|DCer of the Month - March 2011/June 2013/De

    Messages:
    2,499
    Joined:
    Dec 15, 2010
  29. Skillz

    Skillz [H]ard DCOTM Jan,2016

    Messages:
    15,850
    Joined:
    Aug 14, 2004
    I'm gonna try to add mine, feel free to PM me if I f it up.

    edit
    Nevermind, seen it was BigAdvPPD only. :(
     
  30. Linden

    Linden [H]ard|Gawd

    Messages:
    1,140
    Joined:
    Sep 8, 2005
    I have a friend who's folding an 8101 now with an MC 2P. He's reporting that it will miss the preferred deadline. Not sure which CPUs he's running. He's an [H] forums member and has folded for the Horde before. I've asked him to come here and post, as well as FF.
     
  31. 402blownstroker

    402blownstroker [H]ard|DCer of the Month - Nov. 2012

    Messages:
    3,156
    Joined:
    Jan 5, 2006
    It would be interesting to get the machine specs.
     
  32. Punchy

    Punchy [H]Lite

    Messages:
    100
    Joined:
    Sep 4, 2010
    What is this "core hack or other work around" that some or all of your team is using?
     
  33. 402blownstroker

    402blownstroker [H]ard|DCer of the Month - Nov. 2012

    Messages:
    3,156
    Joined:
    Jan 5, 2006
    The answer can be found in the first couple hits of googling:

    "linux corehack script"

    Adding 'overclockers.net' can narrow it down even more :D
     
  34. _k_

    _k_ n00bie

    Messages:
    40
    Joined:
    Sep 28, 2009
    I can't make the preferred deadline with my 2P. 2x6136
     
  35. Schro

    Schro [H]ardness Supreme

    Messages:
    6,786
    Joined:
    Mar 15, 2002
    When I had my 2x 6128, back of the envelope math told me it wouldn't make the 8101 preferred deadline. By at least a day.
     
  36. Patonb

    Patonb Gawd

    Messages:
    608
    Joined:
    Dec 8, 2010
    Judging on 6904 times vs 8101s my dual l5639 won't either by at least a day.
     
  37. Patriot

    Patriot [H]ard|DCer of the Month - March 2011/June 2013/De

    Messages:
    2,499
    Joined:
    Dec 15, 2010
    You are soo close.
    Need to find a way to get you overclocked...
    Is that a 2p or 4p board?
     
  38. tjmagneto

    tjmagneto [H]ard DCOTM x2

    Messages:
    3,007
    Joined:
    Aug 6, 2008
    Be nice. :D
     
  39. _k_

    _k_ n00bie

    Messages:
    40
    Joined:
    Sep 28, 2009
    Asus 2p, Asus KGPE-D16.

    I also have a Tyan board, S8230GM4NR, with 2x6134.
     
  40. Nathan_P

    Nathan_P [H]ard DCOTM x2

    Messages:
    3,077
    Joined:
    Mar 2, 2010
    added some 6904/8101 tpf from my x5670 rig
     
Thread Status:
Not open for further replies.