Slow folding sempron?

botheredbothered Manchester UK
edited February 2005 in Folding@Home
Hi all,
I recently upgraded pc2 with an NF7s and a 2400 Sempron from an Elite mobo and 1.4 Athlon. I thought the Sempron would boost my folding a little but it seems slower, a lot slower. It has done a couple of WUs but seems to be doing them 25%-30% as fast as my other PC. Any tips, ideas?
«1

Comments

  • edcentricedcentric near Milwaukee, Wisconsin Icrontian
    edited February 2005
    Doesn't make sense.
    I do get good production from a 1.4TBird, of course it is clocked at 1.6.
    But the Semp should be noticable faster.
    How much memory? What switches?
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    try tinkers and see if your ppd increase. It may just be the gromacs handicapping you.
  • mmonninmmonnin Centreville, VA
    edited February 2005
    It still shouldne be that bad. The semprons should be a bit better than an AXP, at least they were in other benchmarks.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    well like ed said how much mem?
  • botheredbothered Manchester UK
    edited February 2005
    I'm not sure which switches are set, if any, but the PC has 1Gb of RAM. How do I go about setting a switch and which one?
    edit, Ah, they both have -advmethods set.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    I figured you had -advmethods because of the the wu's you were getting just wasn't sure.

    Let me look into those two proteins for a sec.
  • primesuspectprimesuspect Beepin n' Boopin Detroit, MI Icrontian
    edited February 2005
    maybe you have a process that is hogging CPU time....
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    what speed are you running the 3200 and what speed is the 2400? If stock let me know anyway cause I'm not up on that. Also ...is the sempron cpu folding 100% and is there anything absorbing processes in the background?


    thanks

    edit:\ also what evidence do you have that the sempron is slower than the 1.4? Is the sempron a dedicated folder?
  • botheredbothered Manchester UK
    edited February 2005
    The 3200 is running at 2.19GHz and the Sempron at 1.66GHz, both stock speeds. Task manager on the offending PC shows two entries for folding. FahCore_78.exe , cpu 99% and FAH502-console at 0%.
    Edit. No evidence as such, the Athlon just seemed to do more. It's not a dedicated folder but is on 24\7.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    ok.

    well to get the best assesment of the folding speed you'd need to leave it alone for the duration of an entire frame and you'd get a close estimate.
    I don't run any gromacs on any amd's under 2ghz ...anything less as a rule of thumb I've been running tinkers instead especially if the computer is used often. I have a 2000xp@1.725 and a m2600xp@2.4 and I run only tinkers on both. The only amd I am currently running those 600p gromacs on right now is my home 3000xp@2.4 and it is slow.

    If you're comparing the times between the two processors that you have showing I'm willing to guess that the sempron is folding a 600 pointer and the 3200 is folding something way less. I'm still looking into it though.

    If the 1.4 seemed to do more it's probably because it was folding smaller wu's.
    still just guessing here.

    edit:\\ the 600 pointers are taking 76 to 77 hours on my p4 2.53 & 2.4's respectively.
  • botheredbothered Manchester UK
    edited February 2005
    OK thanks csimon. Which flag do I use to just get tinkers? The 3200 does ok with anything really but maybe the sempron needs just tinkers?
    The sempron is doing a 600 point gromac and the 3200 a 302 point tinker. I wasn't comparing the speed of both, just the sempron as to what it did before.
  • edcentricedcentric near Milwaukee, Wisconsin Icrontian
    edited February 2005
    CS - OK, how in the heck do you get tinkers?
    My old TBird would love them.
  • TimTim Southwest PA Icrontian
    edited February 2005
    Yeah - how can you specify getting just a certain kind of work unit?

    Do you just click the microscope icon with EM3 open and add the commands where it says "Use Extra Launch Commands", check the check box, and have the extra commands typed in that space and save & exit and it'll start doing things with the new comands?

    I have the "-advmethods" and the "-verbosity 9" commands in that box for over a week now, and my BP6 twin CPU system STILL only gets work units between 28 and 47 points. It usually takes andwhere from 30-55 hours to do them. I have no idea what those commands do, I just saw them here and it was suggested to use them.

    Do I even have them in the right place?
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    bothered ...
    remove the -advmethod flag on the sempron box and you will more than likely get tinkers but no guarantee. -advmethods is a mere request for pre-public wu's. just try it on the sempron and see how you like it ...I agree with you about the 3200 I think it's doing fine especially compared to my 2.53g p4.

    ed ...
    remove -advmethods like I said above just remember that you have to stop and start the client once you remove it to get it invoked. also ...finish the gromac then the tinker should come on the next wu dload.

    tim ...
    you can't specify wu's at all. -advmethods is only a request as I said above. no guarantee but the likelyhood of getting gromacs is great for the time being ...that can all change by this afternoon but that's not likely.
    also ...no I don't use emIII to control anything about my f@h ...all I have mine doing is monitoring my processors. In my case that would be a train wreck.
    are you using the gui or console?
  • botheredbothered Manchester UK
    edited February 2005
    This is what I mean. These are both 600 point gromacs but look at the differance in frame time etc. Just seems way too slow.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    bothered wrote:
    This is what I mean. These are both 600 point gromacs but look at the differance in frame time etc. Just seems way too slow.
    I have to agree with you bothered ...I put pencil to paper yesterday when I got home to see my own frame rate on this barton 3000@2.4 and I'm churning these wu's @ 53 hours. Over 200 hours makes no sense really unless there is a deficiency in ram or something I just have no clue.

    Anyone else folding w/ semprons or care to guess?

    It almost seems like the machine is throttled down or something. Can you run cpuz or cpuid to verify the clock speed? Maybe we can have a look all around and figure something out. :thumbsup:
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    aparantly there is much differece between the two cores here bothered. Even though both processors utilize SSE to fold these large wu's the barton core seems to do it far more efficiently for one reason or another. The barton core is an acceptional folder when it comes to these particular wu's and outperforms P4's of equal and some greater frequencies.
    For a more technical explanation I would PM thrax as I'm sure he could satisfy our curiosity no doubt. :thumbsup:

    meantime ...getting tinkers should help offset the handicap somewhat! :fold:
  • ThraxThrax 🐌 Austin, TX Icrontian
    edited February 2005
    I don't see why there should be any difference whatsoever.

    The Barton uses an updated version of the Thoroughbred's SSE engine, and the Semprons are nothing more than a vaguely-reworked Thoroughbred-B.

    My p1134 times on the Tbred-B are ~36 minutes a frame.

    We're missing pieces of the puzzle.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    Thrax wrote:
    We're missing pieces of the puzzle.

    :shakehead something isn't making sense especially since he's showing 100% cpu usage on the client.
  • botheredbothered Manchester UK
    edited February 2005
    Thanks guys, The RAM in here is a lot slower than in the 3200, could it be that? It's the same memory that the Athlon used though.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    IMO that shouldn't make any difference really. Apparantly you have plenty enough ram too.
    The mystery continues.

    Ok another idea can you post your fahlog.txt? ...just the last few lines maybe 9-10 or so should do.
    Also your client.cfg just a cut n paste.
  • botheredbothered Manchester UK
    edited February 2005
    Thanks for your help with this csimon.
    Here's the config.
    [settings]
    username=bothered
    team=93
    asknet=no
    bigpackets=yes
    machineid=1
    local=2

    [http]
    active=no
    host=localhost
    port=8080
    usereg=yes

    And here's the log.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    bothered wrote:
    Thanks for your help with this csimon.
    no problem :thumbsup:
    ok let me see your whole fahlog.txt.

    how do you start and stop the client are you set up as a service?
  • botheredbothered Manchester UK
    edited February 2005
    Oaky doaky. Yes it's set as a service and runns all the time.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    ok.
    look at the log.
    notice when the client is restarted how sometimes it uses Extra SSE boost OK and the frame time is less than one hour per? Other times Extra SSE boost OK isn't used at all and that's when things are slow.

    There's your problem my old friend!

    Without knowing how you stop and restart the client I can't tell you how to fix it but I can tell you that your problem is right there.
    I believe that when you restart the computer if you're set up as a service everything is ok. The problem probably occurs in the way that you terminate the client and restart it otherwise.

    The last two times the client was restarted you will notice this:

    [18:14:35] *
    *
    [18:14:35] Folding@Home Gromacs Core
    [18:14:35] Version 1.70 (October 24, 2004)
    [18:14:35]
    [18:14:35] Preparing to commence simulation
    [18:14:35] - Ensuring status. Please wait.
    [18:14:52] - Looking at optimizations...
    [18:14:52] - Working with standard loops on this execution.
    [18:14:52] - Previous termination of core was improper.
    [18:14:52] - Going to use standard loops.
    [18:14:52] - Files status OK
    [18:15:01] - Expanded 2967231 -> 16166417 (decompressed 544.8 percent)
    [18:15:04]
    [18:15:04] Project: 1141 (Run 71, Clone 6, Gen 6)
    [18:15:04]
    [18:15:07] Entering M.D.
    [18:15:31] (Starting from checkpoint)
    [18:15:31] Protein: p1141_RIBO_FSpeptide_HEL_nospring
    [18:15:31]
    [18:15:31] Writing local files
    [18:15:32] Completed 79647 out of 250000 steps (32)
    [18:32:43] Writing local files <--therefore missing Extra SSE boost OK from here
  • botheredbothered Manchester UK
    edited February 2005
    Ah ha. That PC has all the kids accounts on it and they get swapped quite often. I think it runs on all accounts but there was some confusion when I installed it. I believe the newer folding client lets you run on all accounts with no problem. I'll wait till it's finished this work then install the newer one. Thanks very much csimon. :thumbsup:
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    meanwhile I'd reboot and get Extra SSE boost OK or it will be months! :thumbsup:

    ...and you're very welcome sir!
  • botheredbothered Manchester UK
    edited February 2005
    Found this -
    Use -forceasm. When using EMIII, place that in the "Use Extra Lauch Commands" box.
    Under this thread, How do I keep "Extra SSE boost" always on?

    I have put this in the box in EM3 and rebooted the offending PC. Now it looks like this, much healthier. I've also removed Advmethods, I'll keep my eye on it. Thanks again.
  • csimoncsimon Acadiana Icrontian
    edited February 2005
    ok.
    be sure to update to the newer client when you're done this wu and you shouldn't have to worry about -forceasm anymore.

    -forceasm will ensure that the optimizations are forced upon client restart. When folding tinkers you won't worry about that because they don't use SSE anyway.

    good luck and anymore questions just ask away! :fold:
  • WeedoWeedo New
    edited February 2005
    Uh-Oh!!!
Sign In or Register to comment.