600pt WU Comparison

Omega65Omega65 Philadelphia, Pa
edited October 2005 in Folding@Home
I thought this was interesting. All three have 600pt WUs

FAH1 & FAH2 4400+ 1MB DC @ 2.4ghz HTT240 2x1GB=2GB

FAH3 3500+ Newcastle 512KB SC @ 2.5ghz HTT250 4x512MB=2GB

the 1MB Cache core(s) is folding 43 seconds faster per frame than the 512K core even when it's 100mhz slower......

Comments

  • tmh88tmh88 Pittsburgh / Athens, OH
    edited October 2005
    hmm thats strange

    :cheers:
  • muddocktormuddocktor
    edited October 2005
    No, I've noticed the same thing on some of the advanced wu's. My socket 754 clawhammers outfold my winnie on the 600 pointers and especially the p147x/p1481 series. And my Pentium M desktop system smokes everything in sight on the p147x/p1481 work, averaging them at 900 pts/day folding rate.
  • tmh88tmh88 Pittsburgh / Athens, OH
    edited October 2005
    muddocktor wrote:
    No, I've noticed the same thing on some of the advanced wu's. My socket 754 clawhammers outfold my winnie on the 600 pointers and especially the p147x/p1481 series. And my Pentium M desktop system smokes everything in sight on the p147x/p1481 work, averaging them at 900 pts/day folding rate.

    ok now im confused. so some older technology is better for folding?

    so now bad = good, and good = bad? (for folding)
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    Interesting. I wonder how they compare to the pentiums?
    It would also be interesting to see how they handle double gromacs as well ...they are sending out to a64's aren't they? Probably few and far between perhaps.

    Nice performace on that toldeo core omega! :thumbsup:
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    Here's a better comparison. With my current heatsink combo (the stock 4400+ AMD copper heatpipe) I cant clock it higher or it'll melt (I hit 100C @ 2.6ghz :) ) So I dropped the Newcastle 512K Single core to 10x HTT240 = 2.4ghz

    FAH1 & FAH2 4400+ 1MB (x2) DualCore @ 2.4ghz HTT240 2x1GB=2GB 3-3-3-8 2T

    FAH3 3500+ Newcastle 512KB SingleCore @ 2.4ghz HTT240 4x512MB=2GB 2.5-3-3-8 2T

    The 1MB cache core(s) are 2min per frame faster than the 512K cache CPU. I guess cache matters for the big ones.
  • muddocktormuddocktor
    edited October 2005
    tmh88, it's not the technology per se, it's the amount of L2 cache on the proc that is making the difference. Single core socket 939 processors really don't need all the memory bandwidth that dual channel memory is bringing them, which is why a socket 754 machine with 1 MB L2 cache can perform very close to a socket 939 machine that is using 512 kb L2 cache at the same processor speed. Now when you start talking dual core procs like Omega is using, that extra memory bandwidth is needed with the socket 939 machine to keep from having a bottleneck in the memory bandwidth because you then have 2 separate cores sharing that extra memory bandwidth..
  • tmh88tmh88 Pittsburgh / Athens, OH
    edited October 2005
    muddocktor wrote:
    tmh88, it's not the technology per se, it's the amount of L2 cache on the proc that is making the difference. Single core socket 939 processors really don't need all the memory bandwidth that dual channel memory is bringing them, which is why a socket 754 machine with 1 MB L2 cache can perform very close to a socket 939 machine that is using 512 kb L2 cache at the same processor speed. Now when you start talking dual core procs like Omega is using, that extra memory bandwidth is needed with the socket 939 machine to keep from having a bottleneck in the memory bandwidth because you then have 2 separate cores sharing that extra memory bandwidth..


    ok i get now

    :cheers:
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    muddocktor wrote:
    tmh88, it's not the technology per se, it's the amount of L2 cache on the proc that is making the difference. Single core socket 939 processors really don't need all the memory bandwidth that dual channel memory is bringing them, which is why a socket 754 machine with 1 MB L2 cache can perform very close to a socket 939 machine that is using 512 kb L2 cache at the same processor speed. Now when you start talking dual core procs like Omega is using, that extra memory bandwidth is needed with the socket 939 machine to keep from having a bottleneck in the memory bandwidth because you then have 2 separate cores sharing that extra memory bandwidth..
    Would really be nice to see how the toldeo x2 compares to the manchester x2 ...that would explain how much or whether or not the cache really has everything to do with it. Is there more mem bandwidth with the extra L2?
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    csimon wrote:
    Would really be nice to see how the toldeo x2 compares to the manchester x2 ...that would explain how much or whether or not the cache really has everything to do with it. Is there more mem bandwidth with the extra L2?
    Give me a couple of days...... :)
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    here we go a X2 3800+ (Dual 512K Core) vs X2 4400+ (Dual 1MB Core) @ 2.4ghz

    FAH4 & 5 are the X2 3800+

    600pt WU @ 2.4ghz

    FAH5 X2 3800+ 28:45 per Frame
    FAH1 X2 4400+ 27:01 per Frame
  • muddocktormuddocktor
    edited October 2005
    Looks like the extra cache of the 4400 makes a sizable difference on the 600 pointers then. That's what I figured after seeing how they performed on my socket 754 machines and my socket 939 machine. Both socket 754 machines have clawhammers in them with 1 MB L2 cache and they both outperform my Winnie (which has 512 k L2 cache) in the socket 939 board.
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    Omega65 wrote:
    here we go a X2 3800+ (Dual 512K Core) vs X2 4400+ (Dual 1MB Core) @ 2.4ghz

    FAH4 & 5 are the X2 3800+

    600pt WU @ 2.4ghz

    FAH5 X2 3800+ 28:45 per Frame
    FAH1 X2 4400+ 27:01 per Frame

    wow ...what ram are you running in each? I didn't expect that much difference.
  • muddocktormuddocktor
    edited October 2005
    That's why I'm holding out for either an X2 4400 or a DC Opty 170 or 175. I don't think I want the Opty 165 as it only runs at 1800 MHz and I like the idea of a higher max multiplier with the ram constraints of my EPoX board. It doesn't have the many different ram speeds like a DFI board, just 200, 166 and 133.
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    there's $179 difference between the 3800 & 4400.
  • muddocktormuddocktor
    edited October 2005
    Yeah, I know. :bawling: But I've decided to bite the bullet and spend the extra bux though.
  • tmh88tmh88 Pittsburgh / Athens, OH
    edited October 2005
    I just installed windows x64, will this effect folding at all. Will this make folding faster, or just not effect it at all?
  • muddocktormuddocktor
    edited October 2005
    I don't think it will affect folding much, if at all. I do know that Stanford did some tests to see if there would be any improvement by making a 64 bit client, but the performance difference was too slight for them to mess with right now.
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    csimon wrote:
    wow ...what ram are you running in each? I didn't expect that much difference.
    X2 3800+ 2.4ghz 10x HTT:Mem 240:200 Cas2-3-3-6 1T 2GB (2x 1024MB)
    X2 4400+ 2.4ghz 10x HTT:Mem 240:240 Cas 3-4-4-8 2T 2GB (2x 1024MB)
    A64 3500+ 2.4ghz 10x HTT:Mem 240:240 Cas 3-4-4-8 2T 2GB (4x 512MB)

    the X2 3800+ has different timings than the others but it's frame time is still comparable with my A64 3500+ @ 2.4ghz.

    I think I'm going to change evy box to 2400mhx 10x HTT:Mem 240:200 Cas 3-3-3-8 2T to better comparison
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    ???? :scratch::scratch::scratch:

    X2 3800+ FAH 4 & 5 vs A64 3500+ FAH 3

    X2 3800+ FAH5 29:24 per Frame

    A64 3500+ FAH3 28:54 per Frame

    I think there's some bandwidth contention with the X2 3800+ between the two cores.

    Tomorrow I'm going to replicate the FAH5 WU across all three machines X2 4400+, X2 3800+ & A64 3500+ (all at 2.4ghz) so we can have a true apples to apples test. All machines have 2GB running at 200mhz Cas 3-3-3-8 2T setting

    Stay Tuned......
  • profdlpprofdlp The Holy City Of Westlake, Ohio
    edited October 2005
    Omega, I've been following this thread with quite some interest. If you don't mind me butting in with a little suggestion (more of a request, really), why not compile your research into an article for our FAH page?

    That is, if someone else *cough* Keebs *cough* shorty *cough* prime *cough* doesn't swipe it for the main page first. :vimp:
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited October 2005
    And for perspective (what perspective exactly, I have no clue):
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    muddocktor wrote:
    Yeah, I know. :bawling: But I've decided to bite the bullet and spend the extra bux though.
    Huh I might just join you ...some SPECviewperf scores would surely have me convinced.

    And the decision to way til 1Q06 makes matters even worse ...http://www.digitimes.com/mobos/a20051017A7037.html
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    Ok here we go. I shut down FAH 1-5 and replicated the same WU across all of them (after backing up them up of course :))
    Every system is running 2GB of memory at 200mhz Cas3 3-3-3-8 2T and all CPUs are at 2.4ghz

    X2 4400+ FAH 1 & 2 27:03 per Frame
    A64 3500+ FAH 3 28:58 per frame
    X2 3800+ FAH 4 & 5 30:16 per frame
  • mmonninmmonnin Centreville, VA
    edited October 2005
    Wow, 3 minutes per frame just with the extra cache!
  • Omega65Omega65 Philadelphia, Pa
    edited October 2005
    After seeing that the X2 3800+ is running 78 sec per frame slower than the Clawhammer based 3500+ I shut down the X2 3800+ second core (FAH 5) to see how much of a difference it would make....

    A64 3500+ FAH 3 28:57 per frame
    X2 3800+ FAH 4 29:27 per frame

    The X2 3800+ running only a single core improved by 50 sec per frame but is still 30sec per frame slower than the 3500+ :scratch::scratch::scratch:
  • csimoncsimon Acadiana Icrontian
    edited October 2005
    must be the difference between dedicated and shared cache ...like a pipe of some sort or controller to govern the share ...who knows? :scratch:
  • MedlockMedlock Miramar, Florida Member
    edited October 2005
    Wow. Thanks for taking the time to do all that, Omega. Very interesting. :thumbsup:
Sign In or Register to comment.