forceasm and advmethods with 4.00

kanezfankanezfan sunny south florida Icrontian
edited January 2004 in Folding@Home
what do those two flags do, and do I need to run them on athlon xp based machines running the 4.00 console (not graphical) client?

Comments

  • a2jfreaka2jfreak Houston, TX Member
    edited January 2004
    I think the 4.0 clients use -forcesse rather than -forceasm
    What it does is forces F@H to use SSE instructions to help increase the performance of the processor.

    -advmethods requests gromac units, which are generally worth more ponts and can be crunched more quickly.

    I run them on my AXP machines, but some people have reported issues with -forceasm (and I guess -forcesse might have similar issues) with AXP systems.
  • mmonninmmonnin Centreville, VA
    edited January 2004
    a2j is correct.

    The -advmethods gets you semi-beta WUs which have gone thru internal stanford testing, a beta team test, and is now out for the general public to get. They may not be quite as stable as the ones you can get with no flags tho.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    edited January 2004
    Bartons like both forcesse and forceasm, but client 4.00 is supposed to autodetect asm now and does check validity of SSE also even if force flag is used. With the betas, it was not always doing the forceasm and I have not yet dispensed with it here on either the P4 or the Barton box.

    One side note, if you get big gromacs, look at your core_78 version in log after you start gettign assinged larger gromacs WUs. If, after that, you still have Core_78, you can pause the client and will likely end up with Core_78 of release version 1.55. That version kicks butt with big WUs, it does them in about 90 to 80% of the time the 1.54 core with SAME switches did. Little ones, it is not so much better with, not so much improvement.

    If you get something with an oddball never-seen-this-kind before that might have a number in the mid-to-high 800's or low 900's above 920, your box wants them. they are only 31.4 pointers, but with the new Core release version they go through my Barton at the rate of 12% finished per hour or FASTER-- seen the Barton do 16% of total WU an hour on one of them. I have had ZERO of the new semi-betas crash F@H client or the new core release.

    Way to get new core release:

    Pause or stop client.

    wipe the file Core_78.

    restart client.

    It will very likely get new release of Core_78, which at last I heard was release version 1.55. It was tested with the version 4 client and client betas, very neat release for boht client and the new Core_78.

    I will give you a sample of increase in throughput from the flags being active and the new core release-- The older core did about 3.5 31 pointers in a day. The new one averages 4.2 of same WU grade a day (24 hour time span), and in some cases same set of WUs. The Barton got a bunch of those, I kept logs. Barton old core release version, was running about 10-11 min per percent for Project 1000 WUs when stable and I was not playing with things trying to push the folding. New core release with new client, ALL switches active, about 8.5 to 9.8 min per percent-- varied some by machine load and WU. That took total WU complete time from 17-18 hours total, down to about 9-10 hours for the biggest WUs folding has ever sent out. Those are 50 pointers. Would you rather have actual pointage of about 90 points a day, or 130-140 per day average over long term out of your Barton????

    That is why my two boxes, one P4, one Barton, got 1143 points last 7 days worth of computing (that is understated by stats mirror update lags right now, this week will have about 1260 points if things go smoothly in REAL local processing pointage-- for weekly production). New client, all switches, new core_78 release, rocking points. I OC 12% on Barton box (overall diff in clocking from base), about. CPU is at 2070 actual MHz, has been for 10 days, RAM is only pumped to 172 MHz base from 166 nominal-- an average Barton box, not an extreme box by any means, with GRAPHICAL client running on XP. I have not had to play with voltages much at all.

    John.
  • a2jfreaka2jfreak Houston, TX Member
    edited January 2004
    Can anyone else confirm what Ageek said about v1.55 being 10% (or more) faster than v1.54 of core_78? If so, I'll delete my core_78 because I just checked my logs and I've got v1.54 crunching away on this box and I'm lead to believe it's like that on the other boxes I run too.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    edited January 2004
    Well, if not OCing you will see that much improvement on anything running 1.6 GHZ or above. OR more, if client runs constantly and is stable. IF OCing, try on a box that is not OC'd much first, then phase in, and if you want to keep old client you can do that also by renaming old file to Core_78OLD.bak and that will disable it. OCing is where you get most problems with forceasm switch, and a few foilks who OC have issues with forcesse for some reason not yet a pattern. I crawled the folding community forums before I did this, did this as soon as I knew new client awas out, which was before Christmas, and those figures are averages. BUT, my boxes also ran the betas of version 4.00 client reasonably stably-- all of them.

    If not wanting to jump, WADE first.... :D

    John.
Sign In or Register to comment.