an observation

csimoncsimon Acadiana Icrontian
edited December 2003 in Folding@Home
my 2.5ghz AMD is folding a gromac @ around 6 mins per step.
my 2.5ghz Intel is folding the same gromac @ around 8 mins per step. I can post some logs if you like!

many of you are already aware that amd is the king folder when compared apples to apples or in this case mhz to mhz.

just thought I would renforce that.
«1

Comments

  • mmonninmmonnin Centreville, VA
    edited December 2003
    Pwnage. GO AMD!!
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    those were p1005's ...I'm back to folding the 60 pointers now at about 9 mins per step on AMD
  • a2jfreaka2jfreak Houston, TX Member
    edited December 2003
    Is the Intel an HT processor? How many instances of F@H were running? Were both machines running the same version of F@H? Which flags were used on the clients?
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    The Intel is a P4 non-HT processor 2.53/533 ...1 F@H instance running ...both machines running the FAH 4 beta console prerelease 2 ...both running the core v1.54

    AMD flags = -service -ForceSSE (-"betaflag")
    Intel flags = -service (-"betaflag")

    since I was folding a p1005_ppg10c_pfold on each machine which is a beta wu ...I had an additional flag which I can't disclose here but you know what it is if you are a beta tester.

    Now I am running these flags:
    AMD = -service -advmethods -ForceSSE
    Intel = -service -advmethods

    and folding a p684_TZ2_EXT_EXP
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited December 2003
    For single client configurations, the Barton CPUs are tremendous compared to any Intel CPU. However, running two clients on an Intel hyperthreading-enabled system will outperform a single AMD setup.
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Leonardo

    We are now running the same hardware and I am curious as to what client and flags you are running on your AMD rig. Ihad to lower my OC to 2300 from 2420 because I was having problems and am now on the 3.24 graphical client with no flags. I wasn't sure how much of my problems were due to the OC or the 4.00 Beta I was running.
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    edited December 2003
    Um, my P4, in Linux with a newer kernel and 4.00 Pre2 client and latest Core_78 likes the -forceasm -forceSSE -advmethods switch trio. It is folding at about 90% the effectiveness of the Barton on 98 SE, with DirectX 9.0b in place and stable. The overall prod rates are so close it is not funny-- partly because Linux frees the Linux 4.00 Pre hyperfast after a storage request (file system advantage), but Windows itself, with the latest DirectX, favors AMD of recent genre as DirectX was tuned more for modern games and textures and uses FP\Vector calcs better and the newer WUs that the Pre's tend to get are more heavily vectored (this is why the SSE switch at all). Part of the thing is this-- the Intel is on an Intel chipset board, the AMD is not on an nForce board.

    At a guess, the Intel, with its deeper pipes, does not like Floating Point as well. It was tuned for balance between ALU and FPU, not using FPU heavy optimizations as the AMD chips now are. barton definitely does much better on Folding if both were using Windows, but Linux is Intel skewed more than Windows now is (XP and up) in STOCK and BASE tuning, and takes advantage of Intel balance and feeds the FPU more to apps that need it, and the ALU more to apps that have to have it. Linux base is heavily ALU centric, more FPU open for Folding.

    Yes, you are perfectly correct for Windows alone. Unless you tune Linux from core up for AMD, Intel runs better on it. Windows had to tune more and more for AMD. But part of what you see and I see is not pure Intel vs AMD, it is part the BIAS in the O\S also. Hardware PAIRED with O\S and its ancillary submodules, in other words.

    John.
  • mmonninmmonnin Centreville, VA
    edited December 2003
    The 3.25 client and the new beta clients allow AMD users to force extra optimizations in the client that use assembly optimizations on the CPU. The extra optimizations are 3dnow!, and SSE. With 3.24 all you can get is 3dnow! optimizations but with 3.25 with the -forceasm flag and the new 4.0 clients with the -forceSSE flag you can force SSE optimizations.

    The extra optimizations cause your CPU to become hotter. Everyones does so. Thus most likely causes your instability from too much heat.
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Mine folds @ 47C with the optimizations so I eliminated heat from the picture. I can also run continuous loops of 3D'03 and Prime95 overnight without a problem so was real curious about the issues I was having. Also, where can I get the 3.25 since I don't see it anymore?
  • Straight_ManStraight_Man Geeky, in my own way Naples, FL Icrontian
    edited December 2003
    Yes, mtgoat, I had to put an SLK-900A fan in to get my Beta to run stable-- heat, in this case, partly environmental heat around case, partly warmer air in case, large part CPU is running about 8-9 C under load hotter than not under load with lesser fan and HS, and 4-6 C hotter when under load than not under load with the SLK-900A. I am thinking about an SLK-900U for my Pentium 4 CPU. IT runs 4-5 C higher on average than the Barton does now and is under a lesser overall load than the Barton usually.

    John.
  • fatcatfatcat Mizzou Icrontian
    edited December 2003
    3.25 client...just rename to "FAH3Console.exe"
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Fatcat
    Thanks! ;)

    John
    I'm sorry but you lost me in the translation there.
    When I was running just prior to the problems I my temps were at full load folding with flags on;

    ambient= 24C
    system= 25C
    CPU= 47C

    I had folded earlier in the summer with mcuh higher temps while my A/C was broken and it did just fine.
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    I lowered the vcore from v2.03 to v1.95
    so far:
    system = 28c
    cpu = 50c
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    csimon had this to say
    I lowered the vcore from v2.03 to v1.95
    so far:
    system = 28c
    cpu = 50c
    So what do you you have that Tornado running at? It looks pretty good. :thumbup
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited December 2003
    We are now running the same hardware and I am curious as to what client and flags you are running on your AMD rig.

    MtGoat,

    My AMD rig (@2.3GHz) is running F@H version 3.25 graphical, with flags -advmethods and -forceasm. Given the same Gromacs unit, the Barton rig will complete work units usually about 2 minutes faster than my Intel rig (2800@3500). But, I can run two instances at the same time with the Abit IC7/P4C with no Folding performance degradation versus one instance. Running two console clients, 3.25, with -advmethods.
  • a2jfreaka2jfreak Houston, TX Member
    edited December 2003
    Leo: Are you saying that your p4 @ 3.5GHz will crunch two wus in the same amount of time (+2 minutes) it takes a 2.3GHz Athlon to crunch one? Or are you saying that the p4 averages just 2 minutes more per wu when you take the length of time it took to crunch a wu and then divide by two (assuming each wu took 50% of the CPU) to see how fast it could have been?
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited December 2003
    If the Barton requires X time for a Y Gromacs work unit, the P4C will crunch 2Y(X+2). Add four minutes to a given work unit completed on the AMD, and the Intel will complete two identical work units.

    I'll be happy to check again, but it's been a long time since both computers have been folding the same Gromac units simultaneously. The Intel has been almost exclusively Tinkers for the last couple of weeks.
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Leo,

    Thank you! I will go back to that as that is what I was running before I switched CPU's and it worked well. Then I went to the 4 Beta when I saw it and things went downhill. I know Fatcat posted the console but do you know where I can get the Graphical as I prefer it?

    TIA,
    Larry
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited December 2003
    Larry,

    I also compared graphical client production versus console client production between the two instances running on my Intel. Although the graphical client is purported to consume more resources than the console version, it is by only slightly more. Given the same work unit, usually the graphical client would require about 5 to 20 seconds more per work unit, depending on the complexity of the unit. I wasn't running the screensaver, though.
  • profdlpprofdlp The Holy City Of Westlake, Ohio
    edited December 2003
    mtgoat had this to say
    ...do you know where I can get the Graphical as I prefer it?

    TIA,
    Larry
    Here you go:
  • profdlpprofdlp The Holy City Of Westlake, Ohio
    edited December 2003
    Leonardo had this to say
    ...I also compared graphical client production versus console client production between the two instances running on my Intel. Although the graphical client is purported to consume more resources than the console version, it is by only slightly more.
    Just to echo Leo, late last summer I ran a comparison for a week and found the same thing. About 1% difference between the graphical and console. I have never tried the screensaver, either.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited December 2003
    Hey Larry, sorry for not shooting you the 3.25 graphical installation file. I read your post too fast and your request didn't register with me. Thanks for responding, Prof.
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Thank you both very much!!! :D

    I thought the graphical wasn't too far from the console and I like the ease of use. The screensaver doesn't appeal to me so that's OK. :thumbsup::thumbsup:
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    mtgoat had this to say
    csimon had this to say
    I lowered the vcore from v2.03 to v1.95
    so far:
    system = 28c
    cpu = 50c
    So what do you you have that Tornado running at? It looks pretty good. :thumbup
    tornado is running down to 5000 rpm (w/controller) ...idunno what that would be in cfm but I assume around 50cfm
    I built a buttugly noise baffle that I will take pics of later and post ...it lower the noise down quite a bit to where the tornado can barely be heard at this rpm.
    beyond 5k the pitch becomes unbarable which is part of the reason I retired the 92mm ...and I just couldnt get it to perform as well as the 80mm w/ duct so ...
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    Funny you mention that. I finally got around to trying different fans and casme to the conclusion my original TT SF2 (80mm) is the best fan I currently have on my SLK-900. I tried a Sunon 48 CFM and it is almost as good while both beat my 92mm Enermax Adjustable @ full speed and Panaflow FBA09A12H1A (92mm). So it proves what I was told by others that the 8omm's work better. ;)
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    I think the reason for this is that the 80mm does a better job of concentrating airflow through the fins than the 92mm for the simple fact of it's "tighter" diameter ...maybe a strong 70mm could do even better or even equal ...or a strong 60mm ...I may try the old delta and see what I get but that would ruin the ducting concept as this one only takes 80's and 70's
  • ThraxThrax 🐌 Austin, TX Icrontian
    edited December 2003
    I think I'm going to buy a 92mm fan and make a duct-tape shroud to see what happens.
  • Mt_GoatMt_Goat Head Cheezy Knob Pflugerville (north of Austin) Icrontian
    edited December 2003
    csimon had this to say
    I think the reason for this is that the 80mm does a better job of concentrating airflow through the fins than the 92mm for the simple fact of it's "tighter" diameter ...maybe a strong 70mm could do even better or even equal ...or a strong 60mm ...I may try the old delta and see what I get but that would ruin the ducting concept as this one only takes 80's and 70's
    Yup! I think that is it also. I am still thinking of an aerodynamic adapter like a 120>80 and cutting the flange end off a fan body for mounting puposes.
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    things are too tight in my case to add any sort of flange ...the duct adapter alone barely fits
  • csimoncsimon Acadiana Icrontian
    edited December 2003
    see
Sign In or Register to comment.