new macintosh core - 78

mondimondi Icrontian
edited November 2003 in Folding@Home
Folding Forums

looks like it might finally be worthwhile folding on macs !!!

so hell froze over twice this month

:eek:

Comments

  • mondimondi Icrontian
    edited November 2003
    some preliminary results:

    g4 1 Ghz: Old Core
    [03:39:36] Protein: p912_vill_o
    [03:39:36]
    [03:39:36] Writing local files
    [03:39:53] Testing CPU type...
    [03:39:53] Done testing.
    [03:39:53] Extra AltiVec boost OK.
    [03:39:54] Writing local files
    [03:39:54] Completed 0 out of 250000 steps (0)
    [04:35:15] Writing local files
    [04:35:15] Completed 2500 out of 250000 steps (1)
    [05:30:34] Writing local files
    [05:30:34] Completed 5000 out of 250000 steps (2)
    [06:25:54] Writing local files
    [06:25:54] Completed 7500 out of 250000 steps (3)

    ~ 55 mins a frame

    g4 1 Ghz: new core
    [12:07:05] Completed 87500 out of 250000 steps (35)
    [12:25:15] Writing local files
    [12:25:17] Completed 90000 out of 250000 steps (36)
    [12:44:08] Writing local files
    [12:44:10] Completed 92500 out of 250000 steps (37)
    [13:02:20] Writing local files
    [13:02:22] Completed 95000 out of 250000 steps (38)

    ~ 18 mins a frame

    I have 4 of these running .. as well as 2 dual g5 2Ghz, which are currently finishing off some tinkers ... will post results as i have them

    3x faster on g4 .... and the new core is g5 optimized ... see you in the top 10 soon fella's !!!! :D ..
  • a2jfreaka2jfreak Houston, TX Member
    edited November 2003
    What does the G5 look like?
  • primesuspectprimesuspect Beepin n' Boopin Detroit, MI Icrontian
    edited November 2003
    hehehe good.. I just installed f@h on a g4/800 yesterday...

    heheh look out top 5
  • Mr_BojinglesMr_Bojingles Northern Michigan New
    edited November 2003
    mondi had this to say

    so hell froze over twice this month

    :eek:

    wait, wait, wait....

    alright, here it is :D
  • mmonninmmonnin Centreville, VA
    edited November 2003
    Yeah get some more frame times for us. I would like to know how much it improved.
  • mondimondi Icrontian
    edited November 2003
    heres from all the g4s right now ... trying to get a gro for a g5 but servers are all screwed up:

    1: Project: 912 (Run 0, Clone 21, Gen 21)

    old core:

    [03:39:54] Completed 0 out of 250000 steps (0)
    [04:35:15] Writing local files
    .....
    [12:53:15] Writing local files
    [12:53:15] Completed 25000 out of 250000 steps (10)

    9:13:21

    new core:

    [12:07:05] Completed 87500 out of 250000 steps (35)
    [12:25:15] Writing local files
    ...
    [15:10:05] Writing local files
    [15:10:07] Completed 112500 out of 250000 steps (45)

    3:03:02

    speed up ~ 3x

    2: Project: 344 (Run 14, Clone 41, Gen 1)

    old core:

    [09:35:14] Completed 0 out of 500000 steps (0)
    [10:39:36] Writing local files
    [10:39:36] Completed 5000 out of 500000 steps (1)
    [11:43:57] Writing local files
    [11:43:57] Completed 10000 out of 500000 steps (2)

    2:08:43

    new core:

    [12:27:48] Completed 0 out of 500000 steps (0)
    [12:51:40] Writing local files
    [12:51:42] Completed 5000 out of 500000 steps (1)
    [13:15:35] Writing local files
    [13:15:37] Completed 10000 out of 500000 steps (2)

    0:47:49

    speed up ~2.7x

    3: Project: 724 (Run 3, Clone 28, Gen 6)

    old core:

    [05:47:53] Completed 0 out of 125000 steps (0)
    [06:30:21] Writing local files
    .....
    [09:20:12] Writing local files
    [09:20:12] Completed 6250 out of 125000 steps (5)

    3:32:19

    new core:

    [12:19:05] Completed 12500 out of 125000 steps (10)
    [12:33:43] Writing local files
    .....
    [13:32:25] Writing local files
    [13:32:27] Completed 18750 out of 125000 steps (15)

    0:58:44

    speed up ~3.7x

    4: Project: 344 (Run 51, Clone 76, Gen 2)

    old core:

    [05:01:05] Completed 0 out of 500000 steps (0)
    [06:05:11] Writing local files
    .....
    [15:42:19] Writing local files
    [15:42:19] Completed 50000 out of 500000 steps (10)

    10:41:14

    new core:

    [12:34:02] Completed 150000 out of 500000 steps (30)
    [12:58:36] Writing local files
    ......
    [16:40:06] Writing local files
    [16:40:08] Completed 200000 out of 500000 steps (40)

    3:41:32

    speed up ~2.9x

    roughly translates to about 45 ppd from each g4 as opposed to about 14 ppd before, all in all, pretty dramatic improvement ...

    in fact, seeing as the g5's are running tinkers at a rate of about 35 ppd per proc, the g4s are beating them at 1/2 the clock speed and older architecture... cant wait to get a gro going on the g5s
  • drasnordrasnor Starship Operator Hawthorne, CA Icrontian
    edited November 2003
    Is this just for AltiVec-enabled CPU's or can I expect to see results like these on G3 CPU's? I'm planning to upgrade the dual 180MHz 604e in my S900 to a 1GHz 750FX, since the Sonnet 800MHz G4 upgrade is about as flaky as ihop's pancakes. It'd be nice to know that machine could fold too.

    -drasnor :fold:
  • mondimondi Icrontian
    edited November 2003
    drasnor had this to say
    Is this just for AltiVec-enabled CPU's or can I expect to see results like these on G3 CPU's? I'm planning to upgrade the dual 180MHz 604e in my S900 to a 1GHz 750FX, since the Sonnet 800MHz G4 upgrade is about as flaky as ihop's pancakes. It'd be nice to know that machine could fold too.



    http://forum.folding-community.org/viewtopic.php?t=6121
  • mmonninmmonnin Centreville, VA
    edited November 2003
    Well gromacs uses 3dnow!, SSE, and SSE2 for 386 CPUs and Altivech for Mac CPUs. Since 386 CPUs w/o the above mentioned assembly optimizations do not benefit from the extra code in gromacs I would doubt that any Mac CPU that does not have AltiVec would not benefit either. Not sure what CPUs those are but I hope you see the comparison.
  • drasnordrasnor Starship Operator Hawthorne, CA Icrontian
    edited November 2003
    Just looking at mondi's graph, it looks like the new core Gromacs do alright on the G3's. Thanks for answering that question. 50 minutes a frame for 600MHz 750Cx, so more like 30-40 for a 1000MHz 750FX.

    mmonin: Macintosh PowerPC processors in a nutshell
    IBM PowerPC 601: NuBus Power Macintoshes (G1), 60-75MHz
    IBM PowerPC 603,603e,603ev: NuBus and PCI Power Macintoshes (G1 & G2) 75-275 MHz
    IBM PowerPC 604e: PCI Power Macintoshes (G2) 180-350MHz
    IBM PowerPC 750,750Cx,750FX: Power Macintosh G3, 233MHz-1GHz
    Motorola PowerPC 7400, 7450, 7451: (Altivec) Power Macintosh G4, 350MHz-1.4GHz
    IBM PowerPC 970: (Altivec) Power Macintosh G5, 1.6GHz-2.0GHz

    Notes: 750FX is faster clock per clock than any other PowerPC CPU excepting the 970 on non-Altivec tasks, mostly because it's the only one besides the 970 with on-die full-speed 512KB L2 cache, as opposed to the backside half-speed off-die L2 caches found on other G3's and G4's. A 604e at 180MHz is faster than a 603ev at 250MHz.

    -drasnor :fold:
  • TheSmJTheSmJ
    edited November 2003
    mmonnin had this to say
    Well gromacs uses 3dnow!, SSE, and SSE2 for 386 CPUs and Altivech for Mac CPUs. Since 386 CPUs w/o the above mentioned assembly optimizations do not benefit from the extra code in gromacs I would doubt that any Mac CPU that does not have AltiVec would not benefit either. Not sure what CPUs those are but I hope you see the comparison.

    Actually it's "x86". the 8086, 8088, 286, 386, 486, Pentium, and so on are all a part of the x86 architech.

    [/dork]
  • mondimondi Icrontian
    edited November 2003
    some g5 results, can anyone compare these to a p4 or athlon please ... i cant find any logs of this protein:


    Protein: p254_p341_gnra_den
    [02:46:43] Completed 0 out of 500000 steps (0)
    [02:57:22] Writing local files
    ...
    [07:46:16] Writing local files
    [07:46:18] Completed 140000 out of 500000 steps (28)

    4:59:35


    this is on a g5 2GHz, other proc is doing a tinker @ about 35 ppd

    this roughly works out to 72 ppd (144 if you include both procs) on this protein ... my barton @ 2200 depending on the protein averages around 90 ppd ...

    hmm ... not too shabby, but then of course this machine as configured cost the school about $5500 US ......
Sign In or Register to comment.