Folding hardware question with X2s

2»

Comments

  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2006
    That's strange. I wonder why FAH says that hyperthreading 2 instances 'isn't helpful'? Seems like they should want this, if they get a 125% boost in production per processor.
    That's a generalized, blanket statement. The F@H research is best served by rapid processing, "folding" of protein molecular models, "work units" by clients' computers. Folding two work units on a hyperthreaded P4 requires more time than folding a single WU on the same processor. Stanford has discouraged hyperthreaded folding two instances simultaneously because it slows down the turnaround of assigned-completed work units. I find the argument specious, because if we continue that reasoning to it's logical conclusion, then it must be assumed that the Folding@Home project leaders do not want slower processors of any type, as this would slow down research. Such is not the case! The whole point of the argument, for practical purposes is that we, the volunteer participants, should do whatever we can to ensure timely work unit production and return of data to Stanford. Perhaps one could argue that the earliest, lowest frequency HT P4s really would slow things down too much? But then, why haven't I seen Stanford discouraging processing work units on slow CPUs such as four first generation P4s and 90nm single core AMDs? You also will not find Stanford discourage folding on systems with less than 1GB RAM. Less than 1GB of RAM definitely slows down WU processing on many of the newer, complex Gromacs.
  • EnverexEnverex Worcester, UK Icrontian
    edited October 2006
    the_technocrat wrote:
    That's strange. I wonder why FAH says that hyperthreading 2 instances 'isn't helpful'? Seems like they should want this, if they get a 125% boost in production per processor.

    http://fahwiki.net/index.php/FAH_%26_SMP#Hyper-Threading

    EDIT: that link is why I didn't initially do 2 instances on my HT'ers...:confused2

    It's either running at 125% or a 25% boost, it's not a 125% boost else that would be double and a quarter increase. :wink:
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    so they probably should say '2 instances using HT is ok, but only above 1.xGhz and 1GB+ of memory'

    actually, that is a wiki, maybe it needs to be updated.
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    Enverex wrote:
    It's either running at 125% or a 25% boost, it's not a 125% boost else that would be double and a quarter increase. :wink:

    heh, I meant 125% of per processor output :doh::beer:
  • EnverexEnverex Worcester, UK Icrontian
    edited October 2006
    I think Leo meant in total, so it would be 112.5% per processor :p
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2006
    Most of us who were Folding two instances in hyperthreading were achieving about a 25-30% increase in total production versus Folding one work unit at a time on the same CPU.
    about 25% increased production of running two instances of F@H on on a hyperthreaded P4 versus just one instance
    I wrote what I meant the first time. :D
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    this thread has killed my willingness to understand.

    :P

    I think we've pretty much answered the question to death at this point. :vimp:
  • EnverexEnverex Worcester, UK Icrontian
    edited October 2006
    I wasn't answering a question, I was just correcting you :p
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    to summarize:

    If 'benchmark' =
    1 processor, 1 instance = 100%

    then for other processors of the same speed:
    1 HT processor, 2 instances = about 125% of benchmark
    1 Dual-core processor, 2 instances = about 200% of benchmark
    2 single-core processors, 2 instances = about 200% of benchmark
    2 dual-core processors, 2 instances = about 400% of benchmark
    4 single-core processors, 4 instances = about 400% of benchmark
    4 dual-core processors, 8 instances = about 800% of benchmark

    add to this:
    1 GPU, 1 additional instance = (depends on speed of GPU)
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2006
    Yes!
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    Leonardo wrote:
    Yes!

    :thumbsup:

    I guess eventually we will get a new table for GPU's, but we're going to have to see what pays off more...pixel shaders vs pipelines vs mhz vs ????

    probably some optimal combination

    If I only had my old overclocked 9800PRO...man that was a great card back in the day. :)
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2006
    Presently only ATI has compatible technology for the F@H GPU coding...or vice versa. I don't remember what GPU cores will work with the beta client, but if I remember correctly, it's only cards based on the same core of the X1900~.
  • the_technocratthe_technocrat IC-MotY1 Indy Icrontian
    edited October 2006
    What I mean is, we'll eventually know that it's only helpful to have more than xxxmhz GPU core when you have more than xxx pipelines to feed it or some such spec's. It'ss be a while in beta before we know.

    Butonce we DO know, it'll be great to know how to get the most GPU work done for the $!
  • ketoketo Occupied. Or is it preoccupied? Icrontian
    edited October 2006
    OK i just got a dual core 4600+. OT mildly, but I need a step by step, and I mean detailed step by step, set of instructions - ie., I have no idea how to set up services or 'set affinity', etc - on how to set up 2 instances of F@H. I probably used to but I'm old, y'know?

    I expect this is covered somewhere but a little searching around didn't generate something as detailed as what I need.

    Thankee in advance. And howdy, to many of my old SM associates.
  • ketoketo Occupied. Or is it preoccupied? Icrontian
    edited October 2006
    OK I created 2 folders, put the console version of each. launched 1 and told it to use machine id 1. It downloaded a unit and now is running a loop cycle of closing down the unit then retrying, citing error: Could not get length of results file work/wuresults_06.dat (this number is ever increasing, ie _07.dat, _08.dat, etc.) and Error: Could not read unit 06 file blah blah blah. this is with 5.05 beta, should I try the 5.03 version? Ya, that's what I'm going to try first.
  • EnverexEnverex Worcester, UK Icrontian
    edited October 2006
    Prime was having that error when trying the new GPU folding program I believe, someone gave him a link to what may have been the problem, I'd look at that (in the GPU folding thread). You shouldn't need to set affinity yourself, it should work automatically.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2006
    Another way to set up two instances is, as mentioned above, two separate Folding program Folders. Set each F@H configuration for a different
    "machine ID" number and do NOT set Folding to start automatically with Windows; create a shortcut for each F@H instance and manually move the shortcuts to the "Startup" folder in C:/Documents and Settings\~. I've also installed TrayIt! to minimize the running Windows to the task tray. (TrayIt will minimize any window to the task tray, not just F@H.)

    This is just one way to do it. For me it has worked very well.
Sign In or Register to comment.