A64's and QMD's

muddocktormuddocktor
edited January 2006 in Folding@Home
Since the assignment server is sending out such an extraordinarily crappy selection of crappy WU's to my AMD and Pentium M machines, I decided to do a little experimentation with QMD's with them. I downloaded some QMD work with an extra client on 1 of my P4 machines, then tranferred them over to 1 client on my DC X2 machine and on my P-M laptop and my P-M desktop machine. After getting convergence, all are averaging much more ppd than the crappy Gro and the Tinker WU's I've been getting on these machines. The laptop at 2.0 is averaging 242 ppd, the P-M desktop machine is averaging 347 ppd at around 2300 MHz and most amazingly the client on the A64 machine is averaging 322 ppd without using SSE2 optimizations (at 2433 MHz), since Stanford has issues with using SSE2 on AMD with these WU's (licensing, not operational issues). So if you have a P4 at the house along with a P-M or A64, this is a viable way to increase productivity of your P-M or A64 with the present assignme3nts being handed out by Stanford.

Comments

  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited January 2006
    Hey, that's good news. We were talking about this in another thread, but it didn't dawn on me until now - I don't have advmethods switched on for my P-M laptop. Hmm, I can smell a points production increase.

    Thanks, Mudd.

    EDIT: Arrrghh. I think we just had it in a thread today, but I can't find it -- how/where to attach the -admethods flag to a graphical Folding client?
  • CryptoCrypto W.Sussex UK Member
    edited January 2006
    Leonardo wrote:

    EDIT: Arrrghh. I think we just had it in a thread today, but I can't find it -- how/where to attach the -admethods flag to a graphical Folding client?

    Leo, for the graphical client, I make a shortcut on the desktop. Right click shortcut/properties/target. In the target line you should see something like "C:\Programe Files\Folding@Home\winFAH.exe" -forceasm.
    You can just add -advmethods on the end, make sure there is a space ie ....exe" -forceasm -advmethods.

    Cheers

    Crypto :smiles:
  • DonutDonut Maine New
    edited January 2006
    Muddocktor, thanks for the tip.

    I transferred a QMD to my 242 box, 30min. per frame on a 1912 (a tinker running on the 2nd cpu).:cheers:
  • csimoncsimon Acadiana Icrontian
    edited January 2006
    muddocktor wrote:
    Since the assignment server is sending out such an extraordinarily crappy selection of crappy WU's to my AMD and Pentium M machines, I decided to do a little experimentation with QMD's with them. I downloaded some QMD work with an extra client on 1 of my P4 machines, then tranferred them over to 1 client on my DC X2 machine and on my P-M laptop and my P-M desktop machine. After getting convergence, all are averaging much more ppd than the crappy Gro and the Tinker WU's I've been getting on these machines. The laptop at 2.0 is averaging 242 ppd, the P-M desktop machine is averaging 347 ppd at around 2300 MHz and most amazingly the client on the A64 machine is averaging 322 ppd without using SSE2 optimizations (at 2433 MHz), since Stanford has issues with using SSE2 on AMD with these WU's (licensing, not operational issues). So if you have a P4 at the house along with a P-M or A64, this is a viable way to increase productivity of your P-M or A64 with the present assignme3nts being handed out by Stanford.

    How did you download the QMD's? Sorry I've been out of touch.
  • lemonlimelemonlime Canada Member
    edited January 2006
    I've been getting really crappy ~140 point gromacs on my A64 lately, I've reverted back to timeless tinkers which do really well. 3 minutes per frame on the tinkers, and they finish faster than those tiny gromacs.
  • DonutDonut Maine New
    edited January 2006
    They have to be downloaded on a p4 based machine, then sneaker-netted to an AMD.
  • csimoncsimon Acadiana Icrontian
    edited January 2006
    Donut wrote:
    They have to be downloaded on a p4 based machine, then sneaker-netted to an AMD.
    but how? using what flag?
    I have dual xeon's right here but I'm just getting amber's using -advmethods atm.
  • csimoncsimon Acadiana Icrontian
    edited January 2006
    lemonlime wrote:
    I've been getting really crappy ~140 point gromacs on my A64 lately, I've reverted back to timeless tinkers which do really well. 3 minutes per frame on the tinkers, and they finish faster than those tiny gromacs.
    which tinkers? The 239/244 pointers take me roughly 4 mins.
  • lemonlimelemonlime Canada Member
    edited January 2006
    csimon wrote:
    which tinkers? The 239/244 pointers take me roughly 4 mins.

    Any of the 241/239 pointers take roughly about 3min/frame for me, but this is on my opteron @ 3.2GHz
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited January 2006
    csimon wrote:
    but how? using what flag?
    I have dual xeon's right here but I'm just getting amber's using -advmethods atm.
    Do you have large packets permission set to 'yes'? Heck, I've even got a QMD on my P4M 2.4 laptop (-advmethods, large packets - yes).
  • DonutDonut Maine New
    edited January 2006
    csimon wrote:
    but how? using what flag?
    I have dual xeon's right here but I'm just getting amber's using -advmethods atm.

    the advmethods flag and enable big wu's, On my dual xeons I use the local flag,(force of habit) running 4 units, 2 QMD, 2 reg.
  • csimoncsimon Acadiana Icrontian
    edited January 2006
    lemonlime wrote:
    Any of the 241/239 pointers take roughly about 3min/frame for me, but this is on my opteron @ 3.2GHz

    Well I can't get past 2600. :rant:

    j/k ...I'm happy for you lemon!

    Yep bigpackets = yes and -advmethods.
    I'll wait til these ambers are done and see what I get.

    [settings]
    username=csimon
    team=93
    asknet=no
    machineid=1
    local=150
    bigpackets=yes

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

    [core]
    checkpoint=30
    ignoredeadlines=yes

    [clienttype]
    memory=64
    type=3
    nonet=yes
Sign In or Register to comment.