System Aspects affecting folding...

GHoosdumGHoosdum Icrontian
edited July 2003 in Folding@Home
I know we posted this on the old Icrontic, but I forgot the answers to some of these. I thought it might be a good reference to see how certain CPU aspects affect folding performance. I've added a few more that weren't on my post at ICF... Please respond with the aspect you're talking about and how it affects folding performance. And please add corrections, I've only posted my own poor understanding!

FSB Speed: (to the best of my knowledge, only the total clock speed really affects folding performance... FSB only affects it in its contribution to clock speed)???

System Memory: no effect

3DNow!: adds a moderate performance boost when used, but not as much as SSE.

SSE: adds approximately 30% folding speed increase when used, particularly with Gromacs

L2 Cache: Approx 5% faster completion per WU by doubling L2 Cache: http://short-media.com/forum/showthread.php?s=&threadid=1259

Hyperthreading: Acts as a virtual second CPU, allowing two instances of folding to run, but doesn't give all that much increase over a single CPU of the same total speed???

Comments

  • TheBaronTheBaron Austin, TX
    edited July 2003
    well if system memory doesn't affect folding, then it would make sense that FSB speed wouldn't either, as FSB would directly impact system memory performance
  • shwaipshwaip bluffin' with my muffin Icrontian
    edited July 2003
    I believe that FSB does have some effects on the speed, but not very much.
  • FoldingAddictFoldingAddict Montgomery, AL
    edited July 2003
    FSB has no impact on folding at all. It's the ending Mhz of your CPU that makes the difference. Changing the fsb or multiplier to different values while ending up with the same CPU clock speed will change nothing for folding.

    L2 cache I don't think has much of an impact either, but I'm pretty darn sure that L1 cache has a large impact.

    P4 - 8k L1 cache
    Athlon XP - 128K L1 cache
    P3 - 32k(??) L1 cache

    That just might directly effect how fast a CPU folds, I dunno, the pipelines of the CPUs architecture also has a big role in how fast you'll fold.

    ~FA
  • mmonninmmonnin Centreville, VA
    edited July 2003
    The only things that matter are overal CPU speed, CPU optimizations that are used, and FPU. Thats pretty much it. SD, DDR, or RAMBUS it doesnt matter as long as there is no delay when RAM is used.
  • WuGgaRoOWuGgaRoO Not in the shower Icrontian
    edited July 2003
    mmonnin has a point...i mean cmon..its a goddamn dos box...it couldnt take up any resources as far as ram goes
  • FoldingAddictFoldingAddict Montgomery, AL
    edited July 2003
    Actually the core can take up as much as 30MB of RAM in some cases; the actually console is only a few K though.;)

    ~FA
  • mmonninmmonnin Centreville, VA
    edited July 2003
    I have 25.6kb of mem taken by the core. I was meaning speedwise but 128 will do on any OS for folding.
  • FoldingAddictFoldingAddict Montgomery, AL
    edited July 2003
    I was talking about what WuGgaroo said about it not using much resources.:)

    My Core_ca is using 29MB right now.

    ~FA
  • GHoosdumGHoosdum Icrontian
    edited July 2003
    Now that Omega has run this experiment, we know that L2 cache does help quite a bit... about 5% or so by doubling L2 cache. :fold:

    http://short-media.com/forum/showthread.php?s=&threadid=1259
  • Omega65Omega65 Philadelphia, Pa
    edited July 2003
    it was done totally by accident also. When I switched to v3.25 beta one of F@H2 began dropping WUs, as I didn't have time to fiddle with it, I just dropped the multi back to 11 (I should have just upped the vcore a notch). The Tbred box was running FSB333 11x the whole time.

    Today I noticed all three machines had the same protein (a rare occurance). I took a nap so the machines could fold unmolested and then took some quick screenies.

    So there you are, some (5%) difference but not earth shattering.
  • leishi85leishi85 Grand Rapids, MI Icrontian
    edited July 2003
    FoldingAddict said
    FSB has no impact on folding at all. It's the ending Mhz of your CPU that makes the difference. Changing the fsb or multiplier to different values while ending up with the same CPU clock speed will change nothing for folding.

    L2 cache I don't think has much of an impact either, but I'm pretty darn sure that L1 cache has a large impact.

    P4 - 8k L1 cache
    Athlon XP - 128K L1 cache
    P3 - 32k(??) L1 cache

    That just might directly effect how fast a CPU folds, I dunno, the pipelines of the CPUs architecture also has a big role in how fast you'll fold.

    ~FA

    Athlon XP 64K L1 Cache.
  • mmonninmmonnin Centreville, VA
    edited July 2003
    They all AMDS have 128k since Durons or so.
  • Omega65Omega65 Philadelphia, Pa
    edited July 2003
    mmonnin said
    They all AMDS have 128k since Durons or so.

    Ever since the original K7 500mhz Slot A
  • mmonninmmonnin Centreville, VA
    edited July 2003
    OK all K7s then.
Sign In or Register to comment.