Running two SMP Clients.. does HT do anything?

TrumandrummerTrumandrummer Taylor Michigan Icrontian
edited September 2009 in Folding@Home
I finally set up fedora 11 in virtual box.
Right now im running Vista with a GPU system tray client. Then I have Virtual box running fedora 11 with two smp clients.

I have HyperThreading turned off on my cpu. Will this effect the SMP clients at all?

Thanks a lot

EDIT: I also notice that my cpu usage is way down. When I am running the GPU and two smp clients in Vista, my cpu usage hits just above 50%.....
Right now, running a GPU and then two smp clients in Linux, I am only using 30% cpu.
hmmm, I wonder if thats a good thing or if it means the Linux clients are not pushing as fast. They do not seem to be moving very fast at all and It keeps returning "timered checkpoint triggered"

Comments

  • QCHQCH Ancient Guru Chicago Area - USA Icrontian
    edited August 2009
    WOW... that's a ton of Folding. I think most here would say NOT to be running all those at once. Most using quad core CPU's and a nice GPU... they run one SMP and one GPU. The SMP will use 2 CPU's and the GPU kinda take one more. That leaves one free. I am not up to date on SMP/ GPU so I'm not sure the optimal setup.

    As for Hyper Threading... NO!!! It just pretends to split each CPU into 2 virtual CPUs. HT works for simple apps but not Folding.
  • SnarkasmSnarkasm Madison, WI Icrontian
    edited August 2009
    Q, with the newer drivers, the GPU executables use hardly any CPU cycles - it's all contained on the GPU. That frees up a four-core machine to use two SMP clients, one for every two cores.

    That said, I don't know about the HT question, so I can't answer it. Just wanted to point out the good reasoning for running double SMP clients.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Well, I downloaded FahMon,
    My SMP clients are only pulling 370 PPD each. So 740 PPD
    That seems really low. Something has to be wrong
    I see posts online from people with i7 920's getting 2,000+ PPD. Some claim they didn't even over clock.
  • lordbeanlordbean Ontario, Canada
    edited August 2009
    that is very low PPD for an i7... my Q9450 @ 3.46GHz gets between 2.3 and 3k per day, depending on the work unit...

    I don't know linux at all so I have no suggestions, but I can confirm that is not a normal PPD range for a core i7.
  • SnarkasmSnarkasm Madison, WI Icrontian
    edited August 2009
    Are you running it with the -smp flag set in the arguments?
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    lordbean wrote:
    that is very low PPD for an i7... my Q9450 @ 3.46GHz gets between 2.3 and 3k per day, depending on the work unit...

    I don't know linux at all so I have no suggestions, but I can confirm that is not a normal PPD range for a core i7.

    Actually I shut off the linux smp clients because they were not working right. The 370 PPD each, is with the two SMP clients running in Vista.



    Snarkasm,
    No I wasn't running them with the -smp flag. How do I try that?
    I know in Linux I can just issue the command ./fah -smp
    But right now I got the two SMP clients running in Vista.

    My actual folding in the past 24hours is over 10,000. But this is only because I have a GPU client set up on my dads 8800gts that is pulling in over 2,000 PPD. My SMP folding is holding back my score a bunch.
  • _k_k P-Town, Texas Icrontian
    edited August 2009
    create a shortcut to launch the SMP clients and open up the properties on the shortcuts. Add -smp to the end of the Target location line. i.e. "C:\Program Files (x86)\Folding@Home Windows SMP Client V1.01\Folding@home-Win32-x86.exe" -smp
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    _k_ wrote:
    create a shortcut to launch the SMP clients and open up the properties on the shortcuts. Add -smp to the end of the Target location line. i.e. "C:\Program Files (x86)\Folding@Home Windows SMP Client V1.01\Folding@home-Win32-x86.exe" -smp

    Ok thanks K :)

    It comes up and says 4 cores detected. Then it says "unable to connect to Nick PC: 8676. Sock error ......... " it closes itself off before I can get the rest.

    Yea, it should be fine. I do not know what that error message means. My firewall is not blocking it, and neither is my router blocking that port.

    EDIT: here is a snapshot of the error

    errorknp.jpg
  • _k_k P-Town, Texas Icrontian
    edited August 2009
    Make sure you running everything in administrator since you are vista, try reinstalling. Are you using the systray or console version, most of use the console and use Trayit to put it in the system tray. But you have to start the install from the cmd window as admin, just follow the folding at home guide if you install it.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    _k_ wrote:
    Make sure you running everything in administrator since you are vista, try reinstalling. Are you using the systray or console version, most of use the console and use Trayit to put it in the system tray. But you have to start the install from the cmd window as admin, just follow the folding at home guide if you install it.

    Ok, thanks
    Thats how I ran it, was as administrator.
    I just have the console version.
    Ill try to re-install them. I may not have installed them as admin though.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Ok. I re-installed one of the SMP clients. It seems to be working with the -smp flag given.
    Hopefully I will some better PPD now. I will report back.
    Thanks a lot

    One thing that I noticed. With just the one re-installed SMP. My cpu level is at 100%. It only used to be at 50%
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Yep that seems to have fixed it. FahMon is reporting just one SMP client at 3800 PPD.
    Awesome! I am going to set up the second SMP client in a little while.
    Thanks a lot everyone!
  • SnarkasmSnarkasm Madison, WI Icrontian
    edited August 2009
    Similarly, instead of using the shortcut trick, if you just watch for the spot in the configuration where it asks if you want to add any arguments when you run it, you can add smp in there and it'll run it with the smp flag without any extra tricks.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Snarkasm wrote:
    Similarly, instead of using the shortcut trick, if you just watch for the spot in the configuration where it asks if you want to add any arguments when you run it, you can add smp in there and it'll run it with the smp flag without any extra tricks.

    Oooo duh. I just re-set them up too. I didn't even think about the arguments spot.

    Well When I have 1 smp client running I get about 3800 PPD
    When I have two running, each client is only pulling like 1600 PPD

    Running one seems to be getting me better performance.
  • SnarkasmSnarkasm Madison, WI Icrontian
    edited August 2009
    It will vary depending on what work units you get, too, don't forget. But yeah, it'll only give you a boost in a few situations.
  • makoa26makoa26 Las Vegas
    edited August 2009
    You can also add flags to your config file located in your smp directory. If you want to run two smp clients try using Nick 4 Evas Affinity changer. http://distributed.org.ua/forum/index.php?showtopic=1149 It should give you a ppd boost with 2 smp clients. As for hyper threading, the way I understand it is that the client uses certain parts of the execution core only. Hyper threading is a way to keep the entire execution core busy. For example if thread 1 was using ALU, then thread 2 would be able to use FPU. Since both folding threads would be using the same portion of the core, then they would be "competing" for the same resources.........I could be wrong though.
  • lordbeanlordbean Ontario, Canada
    edited August 2009
    Oooo duh. I just re-set them up too. I didn't even think about the arguments spot.

    Well When I have 1 smp client running I get about 3800 PPD
    When I have two running, each client is only pulling like 1600 PPD

    Running one seems to be getting me better performance.

    I'd like to note, this is the same behavior that I got from one vs two SMP clients in Vista. My Q9450 pulls higher PPD running a single client.

    I'm still not convinced running 2 SMP clients is at all useful. You complete 2 work units in twice the time... there's no gain.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Yea. For now I am going to run one SMP client. I am going to try what makoa26 said.

    Do you guys think that 72° C is a safe temperature for the Processor?
    The program that came with my computer is reading that it is only at 56º C. But I downloaded RealTemp, and it is at 72º C while running my SMP ( + or - a few degrees) No matter what clock speed I put it at, it does not seem to change. If I have it set to 3.8 ghz, it is the same temperature as 3.2 ghz while I am running my SMP client.
  • ardichokeardichoke Icrontian
    edited August 2009
    Technically speaking, from a programmers standpoint anyway, an SMP client should make use of all available cores, not just two. Thus the behavior that Truman and bean are seeing would be correct, in theory at least (that is, running 2 SMP clients gives no PPD gain, and perhaps a slight loss, over 1 SMP client). Then again, I'm not familiar with the internals of the folding software. Writing an SMP client and limiting it to 2 cores though is pretty pointless, it makes much more sense that they would write it to detect how many cores there are and launch an appropriate number of threads/processes.
  • _k_k P-Town, Texas Icrontian
    edited August 2009
    The whole reason for running two is that Windows does not assign core affinity or allocate threads 100% as it might be reporting and the second client is an attempt to be closer to 100% real workload on all cores. I have seen a loss and gain in ppd in different WUs what really needs to be done is a study of several different WUs, point values and variants within the same point value, while using a single and dual SMP. Reasons like these are why some people run linux in vm, they take the hit in CPU load overhead to get the better core utilization.
  • ardichokeardichoke Icrontian
    edited August 2009
    What I'd like to see is a study as to weather the overhead of running a VM offsets the gain of running folding in Linux. I'm pretty sure the best option is to just run Linux on as many of your folding boxes as possible :-D
  • SnarkasmSnarkasm Madison, WI Icrontian
    edited August 2009
    The benefit to folding under Linux is significant enough that even with the overhead of a VM, you still end up with higher output than running a Windows SMP client.

    Yes, running in native Linux is better than running in native Windows, but running in VM Linux is better than running in native Windows, and you still get to have Windows around for when you need it.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited August 2009
    Snarkasm wrote:
    The benefit to folding under Linux is significant enough that even with the overhead of a VM, you still end up with higher output than running a Windows SMP client.

    Yes, running in native Linux is better than running in native Windows, but running in VM Linux is better than running in native Windows, and you still get to have Windows around for when you need it.

    Yea see thats my problem. I would have no problem just folding in Linux (not VM). But I cant seem to get the GPU client set-up in Linux. Under wine or anything. So I just decided to try VM. I followed the install instructions for Linux SMP, but something must have went wrong. Because it was running really slow, and it was not using 100% of my CPU like I had configured it too. Could have been something to do with Virtual Box and not so much Linux. :scratch:
  • makoa26makoa26 Las Vegas
    edited September 2009
    I think you hit it right on the nose. I don't think virtualbox uses Vt extensions. However, if you want to give folding in a VM another shot, I suggest checking out this post http://www.maximumpc.com/forums/viewtopic.php?p=946781#946781. I tried it out myself when I first read that guide. I got a massive ppd boost........when my rig didn't crash. It seems that vmware folding doesn't play nice with my ATI folding. Maybe nVidia folding fares better. I still have the installer for VMware player if you don't want to "sign up" on VMware's site.
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited September 2009
    makoa26 wrote:
    I think you hit it right on the nose. I don't think virtualbox uses Vt extensions. However, if you want to give folding in a VM another shot, I suggest checking out this post http://www.maximumpc.com/forums/viewtopic.php?p=946781#946781. I tried it out myself when I first read that guide. I got a massive ppd boost........when my rig didn't crash. It seems that vmware folding doesn't play nice with my ATI folding. Maybe nVidia folding fares better. I still have the installer for VMware player if you don't want to "sign up" on VMware's site.

    Nice find. I will for sure give this a shot when I have the time.

    Im already registered on vmwares site, so that should not be a problem.

    Thanks a lot :)
  • TrumandrummerTrumandrummer Taylor Michigan Icrontian
    edited September 2009
    Sadly, I will have to shut off my SMP client today. It is gonna lower my points but whatever. The last like week here has been cold.... and I have had my window open. Well today is a little more hot... and my CPU is going over 80 degrees.
    I ordered at good heatsink... but it hasn't come yet.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited September 2009
    Q, with the newer drivers, the GPU executables use hardly any CPU cycles - it's all contained on the GPU.That frees up a four-core machine to use two SMP clients, one for every two cores.
    Yup, exactly. I have four machines folding, all running multiple CPU + multiple GPU clients. Two of the boxes each run 2 X SMP + 4 X GPU clients. Four GPU clients require about 1% of GPU processing power.

    A poster above remarked the he thought 2 X SMP clients run on a CPU took twice as much time as 1 X SMP. Hmm, I don't know about Vista. My rigs are all WinXP. I get about 50% better production running 2 X SMP on a quad-core (Q6600) as I do a single SMP client.

    I wish to note my comments all pertain to Windows XP. I have no VMWare or VirtualBox instances.
  • lordbeanlordbean Ontario, Canada
    edited September 2009
    Leonardo, that's been the staple of my argument, essentially. I believe you all when you say running 2 SMP clients helps, but everyone so far who has said that, runs windows XP. I run Windows Vista, and it's entirely possible the behavior I observe is because Vista's SMP-handling code is better than XP's.
  • LeonardoLeonardo Wake up and smell the glaciers Eagle River, Alaska Icrontian
    edited September 2009
    Vista's SMP-handling code is better than XP's
    That wouldn't surprise me.
Sign In or Register to comment.