Crappy Folding Performance - Please Help!

[Clocker]

Help!

My Folding performance sucks. I'm not sure why it sucks but it sucks.

I don't think there are any issues with my single CPU P3 machine. I think my DUal Athlon box is giving me some trouble though. It seems as though, even though I'm running two instances of F@H, my performance is no better than that of a single CPU machine. Also, the points I have been getting per unit of work seem really low (LESS THAN ONE per work unit) which leads me to believe each processor in my dual CPU system (Athlon XP 1700+) is processing work units slower than it should be. Could they be interfering with eachother?

I'm not sure what is causing it but I have compiled some screen shots of how my system is configured. If someone would look at what I've got and make any suggestions, I'd really appreciate it!

The screen caps are here:

http://www.storageforum.net/clocker/folding_problem/

Thanks,
Clocker

[Clocker]

My UserName and UserID for each installation on my Dual Athlon Machine are identical.

THe The USerName on my P3 machine is also identical to that of the Dual Athlon machine but the User ID is different. Do I need a different UserID for the 2nd F@H instance on my Dual rig? How can I get one if that is the case?

I'll be researching it but let me know if you know...

Thanks,
Clocker

[Clocker]

I just found out that each instance of the client must have a separate UserID. I installed it separately into different directories but each installation has the same UserID.

How can I get a new User ID??

C

[Bartender]

Clocker, how did you get the two instances to run? On my Dual PIII I installed the software twice in two different directories, and then launched each one separately. However, when I launched the second executable, I got a message telling me that the program was already running. One thing I did notice with the dual system, was that with one instance running, my CPU utilization was between 50-80% (this us under 2000 Professional).

[me]

From http://folding.stanford.edu/faq.html


Does Folding@home run on dual processor machines?

Yes, with the console (text-only, no-nonsense) version. The best way to do this now is to install F@H twice in two (or four, however many processors you have) DIFFERENT directories, and run them with the flag "-local". If two copies of F@H run in the same directory, they will interfere with each other, and much badness will occur. You should actually be able to run one graphical F@H and the rest text-only F@H copies, as long as the graphical one is in the default location and the others are in custom locations.

We have been working on a multiple processor version with parallelized code using OpenMP. However, we have discovered that OpenMP parallelized code does not scale very well at all with processor number with our code. So we are investigating other options for parallelization, and a final date for a parallelized version is uncertain at this time.


Are you using the "-local" switch?

[P5-133XL]

Sorry but I think that everything is setup properly and working properly. Your problem is that the folding project you are assigned is NativeBBA-300K. If you look at the Project lists (folding.stanford.edu/psummary.html) you will find that NativeBBA5-300K gives .6 points per WU. My Project is 601 ProteinA and it gives 5 points/WU and that is why I have 15 points for 3 WU's and you are generating 21 WU's and only geting 20.37 points. I also not that it takes my processor (Athlon MP 1900+) 8 minutes per frame vs. your 1.5 minutes per frame. So in theory you should finish far more frames, but they are worth far less.

Count your blessings; At least you aren't doing project 413 = TRP SIDECHAIN - OPLSAA with give 0 points per WU.

Now the question is, how does Stanford decide who gets what project?

[me]

Now the question is, how does Stanford decide who gets what project?

That's simple. They give all the lousy ones to Clocker. And Tea. :wink:

[Clocker]

I'm getting closer here guys. Thanks for you input.

Here is what I have found....

WHen I launch FAH3Console.exe from within each individual directoy manually (within file manager) a unique User ID gets assigned to each instance of the client running (from the individual directories).

However, when I start the client as a service, each instance has the SAME UserID.

I AM using the -local switch (correctly I hope!) but I'm using it with the FAH3Console.exe executable. Which executable does the -local switch work with? Here is how the service is configured in the registery...notice the -local swith . Is that correct??

http://www.storageforum.net/clocker/folding_problem/registery.gif

[Mercutio]

Since you have two instances of F@H running, have you set processor affinity for each to only one of your CPUs? Might help.

[Clocker]

It turns out that the UserIDs are the same for each FAH3 instance because the -local swith is not working with FAH3.

The -local switch works fine with v2.19 so I'm going to use that. SMP users should not use FAH3 until that gets fixed!!

Merc- I've looked into playing with CPU affinity but could not figure out how. ANy hints? Personally, I don't think it will make a difference here but I would like to know how to do it.....

Thanks!

C

[NRG = mc²]

To set the affinity, right click the process in task manager and there is an affinity setting in the menu... set one to CPU0 and the other process to CPU1.

[Clocker]

Hmmm..I can do as you describe for some processes but not the FAH processes. Must be a bug with the FAH program....

Thanks nRG!

C

[P5-133XL]

I can set the affinity settings to my FAH processes.

[Clocker]

Mark-

Which version of FAH are you running? Whih process are you setting affinity for?

Thanks,
C

[P5-133XL]

The FAH version is Console 3.0 beta 1.
The process is FaHCore_65.exe

[Prof.Wizard]

Says all P5-133XL,
I believe Clocker is using v2.19, right?

[Clocker]

Says all P5-133XL,
I believe Clocker is using v2.19, right?

COrrect...

C

[P5-133XL]

I don't understand. The gif's clocker supplied at the start of this thread indicates he is using console 3.0 beta 1 just like me. (look at the last gif)

[Clocker]

I don't understand. The gif's clocker supplied at the start of this thread indicates he is using console 3.0 beta 1 just like me. (look at the last gif)

Mark-

I switched to 2.19 after I found out that the -local switch does NOT work with any versin of FAH3 at the moment. Because the -local switch is not working, only one instancane of the FAH3 software can be run on an SMP rig or they will conflict invalidating the results for one (or both!). SO, I swithed to the 2.19 software so I could run two instances of FAH on my MPX rig. I wanted to set affinity for each of the two processees but apparently (with the 2.19 engine) I can't do that...

Thanks,
C