Pentium 4 with Hyper Threading

ThelemechThelemech Victoria Icrontian
edited November 2007 in Folding@Home
I just installed fold@home on a P4 630 HT 3.0 Ghz 2 GB Ram and I am going to ask a question that has probably been asked several times before on processors with HT; but how do you get "both" processors to fold ...currently only half(50%) is folding. Install and run two instances of the program?

Comments

  • ThraxThrax 🐌 Austin, TX Icrontian
    edited October 2007
    Use the SMP client.
  • primesuspectprimesuspect Beepin n' Boopin Detroit, MI Icrontian
    edited October 2007
    The SMP client will be terrible on that processor, don't you think?

    I'm not Pentium 4-saavy enough to answer this question accurately, but I think running two instances of the normal client will be far more effective on that CPU than the SMP client.
  • ThraxThrax 🐌 Austin, TX Icrontian
    edited October 2007
    It's not as good as a full dual-core CPU, no, but neither is HT.
  • QCHQCH Ancient Guru Chicago Area - USA Icrontian
    edited October 2007
    HT is synthetically splinting the CPU into two cores and SMP likes multiple cores. The problem is that SMP would kill the HT P4. I'd go with two instances of Folding...
    Mt_Goat wrote:
    Just create a folder for F@H then inside it create a folder named F@H1 and another named F@H2. Then put one exe of the console version in each but don't start yet. Then go back and start each one and set each as a service. Set the one in folder 1 as machine id =1 and folder 2 as machine id =2. It is that simple! :D
  • GHoosdumGHoosdum Icrontian
    edited October 2007
    What Q said.
  • ThelemechThelemech Victoria Icrontian
    edited October 2007
    QCH2002 wrote:
    HT is synthetically splinting the CPU into two cores and SMP likes multiple cores. The problem is that SMP would kill the HT P4. I'd go with two instances of Folding...

    Thanks Q for the info from Mt Goat!

    What a difference between a p2 400 MHz 320 Ram and the P4!!! It was going to take 3 months for the p2 ; 2 to 4 days for the P4!!! :tongue::cool2:

    ....I cannot wait to see the difference with the Quad 6600 :wow2:
  • QeldromaQeldroma Arid ZoneAh Member
    edited October 2007
    Thelemech wrote:
    ....I cannot wait to see the difference with the Quad 6600 :wow2:


    As far as FAH is concerned, the difference isn't quite as dramatic- But dramatic enough.

    Below is a comparison of a single 2.8GHz P4 Hyperthread and a stock Q6600 Core working on the same WU (the Q6600 actually finished the same WU the P4 started). According to this, a Q6600 core has a raw cycle-to-cycle computational improvement of about 3.5 times that of an equivalent P4C Hyper-thread.

    2.8GHz P4C HT2
    [04:06:58] Protein: p2423_Ribo_asp_aux
    [04:06:58]
    [04:06:58] Writing local files
    [04:06:58] Completed 157290 out of 250000 steps (63)
    [04:07:10] Extra SSE boost OK.
    [04:14:55] Writing local files
    [04:15:02] Completed 157500 out of 250000 steps (63)
    [05:22:28] Writing local files
    [05:22:28] Completed 160000 out of 250000 steps (64)
    [06:29:12] Writing local files

    2.4GHz Q6600 Core2
    05:05:51] Protein: p2423_Ribo_asp_aux
    [05:05:51]
    [05:05:51] Writing local files
    [05:05:51] Completed 186790 out of 250000 steps (75)
    [05:05:53] Extra SSE boost OK.
    [05:11:43] Writing local files
    [05:11:43] Completed 187500 out of 250000 steps (75)
    [05:32:20] Writing local files
    [05:32:20] Completed 190000 out of 250000 steps (76)
    [05:53:26] Writing local files
    [05:53:26] Completed 192500 out of 250000 steps (77)
  • ThelemechThelemech Victoria Icrontian
    edited October 2007
    That is still a sizable difference, well worth my future investment of a Quad 6600 system. And of course I plan on overclocking as much as she can handle.
  • yaggayagga Havn't you heard? ... New
    edited November 2007
    Qeldroma wrote:
    As far as FAH is concerned, the difference isn't quite as dramatic- But dramatic enough.

    Below is a comparison of a single 2.8GHz P4 Hyperthread and a stock Q6600 Core working on the same WU (the Q6600 actually finished the same WU the P4 started). According to this, a Q6600 core has a raw cycle-to-cycle computational improvement of about 3.5 times that of an equivalent P4C Hyper-thread.

    2.8GHz P4C HT2
    [04:06:58] Protein: p2423_Ribo_asp_aux
    [04:06:58]
    [04:06:58] Writing local files
    [04:06:58] Completed 157290 out of 250000 steps (63)
    [04:07:10] Extra SSE boost OK.
    [04:14:55] Writing local files
    [04:15:02] Completed 157500 out of 250000 steps (63)
    [05:22:28] Writing local files
    [05:22:28] Completed 160000 out of 250000 steps (64)
    [06:29:12] Writing local files

    2.4GHz Q6600 Core2
    05:05:51] Protein: p2423_Ribo_asp_aux
    [05:05:51]
    [05:05:51] Writing local files
    [05:05:51] Completed 186790 out of 250000 steps (75)
    [05:05:53] Extra SSE boost OK.
    [05:11:43] Writing local files
    [05:11:43] Completed 187500 out of 250000 steps (75)
    [05:32:20] Writing local files
    [05:32:20] Completed 190000 out of 250000 steps (76)
    [05:53:26] Writing local files
    [05:53:26] Completed 192500 out of 250000 steps (77)

    Hey, that looks just about exactly like my 2.8c. Of course that is 1/2 instance of the 2.8 and 1/4 of the q6600 not considering smp correct?
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited November 2007
    When I ran HT enabled processors - overclocked 2.8 and 3.0 Northwoods, I ran two instances of Folding@Home. It worked very well. The production was about 25-30% greater than one instance folding alone.

    SMP? Hmm, I don't think the a single CPU, HT or not, will complete work units on time. It might be worth a try.
  • KentigernKentigern Milton Keynes UK
    edited November 2007
    My P4 3.2 ghz HT couldn't cope with the SMP's so went back to 2 x FAH achieved 50k in a little over a year.
  • SPIKE09SPIKE09 Scatland
    edited November 2007
    Leonardo wrote:
    When I ran HT enabled processors - overclocked 2.8 and 3.0 Northwoods, I ran two instances of Folding@Home. It worked very well. The production was about 25-30% greater than one instance folding alone.

    SMP? Hmm, I don't think the a single CPU, HT or not, will complete work units on time. It might be worth a try.
    It really isn't worth it it really slows down the science, even if they make the deadline it will only be the final deadline not the preferred deadline.
  • QeldromaQeldroma Arid ZoneAh Member
    edited November 2007
    yagga wrote:
    Hey, that looks just about exactly like my 2.8c. Of course that is 1/2 instance of the 2.8 and 1/4 of the q6600 not considering smp correct?

    Yep- so if you look at it on a total CPU-to-CPU basis, you're picking up about 7 times more folding power with the Quad on this WU.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited November 2007
    it really isn't worth it it really slows down the science
    I never bought that argument. If we were to extend that logic out a bit, then Stanford would discourage the use of any computer with a CPU less than Core 2 Duo or Core 2 Quad. An HT enable CPU can complete two work untis FASTER than the same CPU working on the same work units one after the other.
  • SPIKE09SPIKE09 Scatland
    edited November 2007
    Leonardo wrote:
    I never bought that argument. If we were to extend that logic out a bit, then Stanford would discourage the use of any computer with a CPU less than Core 2 Duo or Core 2 Quad. An HT enable CPU can complete two work untis FASTER than the same CPU working on the same work units one after the other.
    If it was better for the project they would encourage it surely, getting a wu back in 1 day rather than say 2.5 days over a projects that has 400 generations, potential delay 1.5 x 400 eg 600 day potential delay.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited November 2007
    In a given time period, an HT CPU running two instances will return more completed work units than the same CPU running only one client. Stanford gets more work units returned. I just don't see that they'd rather have fewer work units returned faster.

    Again, if delay is a problem, then why do they accept work from older machines at all? It's a logical inconsistency.
  • SPIKE09SPIKE09 Scatland
    edited November 2007
    You expect logic from the mad scientists Leo, biology is intrinsically illogical at times add some computers and well your guess is as good as mine. As for supporting old hardware well, plenty folks still running it and they do the leg work on the less time sensitive projects. I think ver6 or 6.1 will control wu distribution better, interesting debate in the end all is good as long as it folds.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited November 2007
    interesting debate in the end all is good as long as it folds
    No argument from me on that.
Sign In or Register to comment.