Wildly varrying TPFs

[Quisarious]

First I have to say thanks for all the great info I've been able to gather from this site. Given the incredible knowledge about folding, I was hoping someone might be able to help me out here...

I installed FAH V7 Client on my workstation (dual Xeon E5-2670's) a couple of weeks ago and have been getting TPFs all over the place.

On 6903's my TPF varies between 15:21 and 25:20 (over 3 WU's mean was 18:22, SD 2:41). I have used the computer off and on while it's folding, but only for browsing/email, and the TPF variance is essentially the same as when left completely alone.

I'm using the V7 client/control modules, and have a fairly fresh install of Ubuntu 12.04. I installed the kraken on the a5 core, and confirmed it's working (via top), though saw no significant change in TPF times. CPU utilization was pegged at 3180+ before and after installing kraken.

I'm don't really care point wise, but I'd like to be able to help as much as possible, and it seems the set up is capable of more.

Any thoughts would be greatly appreciated!

Thanks.

 

[402blownstroker]

No kraken + other usage = lower TPF

Sometimes when I process large collections of video content with single thread ffmpeg running taking up 1 core, TPF often increase 50% to 100%.

I have learned to live with it :shrug:

 

[Quisarious]

Thanks for the reply, but I do have the kraken wrapper installed, though it makes no difference for me at this point. And even with zero use, I get the same TPF variation.

It's also not a throttling issue, temps are good and all the cores are staying at appropriate turbo speeds throughout.

 

[Mr. Pedantic]

Are you using the computer for anything else at all? Because of the bonus and because of the way it's been coded, the F@H cores are really bad at dealing with even a small amount of usage. All the threads have to be in sync, and for some reason I think it's been coded so that as many threads as possible work at 100% and the last thread(s) struggle on their own, down to 0% usage at times. This means that if you peg 1 complete core of your system, if it weren't for the fact that OS' reallocate threads to cores periodically, you would never get a frame done (note that this hasn't been confirmed, it's just my observation plus what I think may plausibly be happening). Even if you use 10-20% of a core, PPD drops precipitously.

 

[Quisarious]

during the day I occiasionally use the browser or check email, but I've also left it alone for 5 days last week and the variance is essentially the same.

As an example, I hadn't touched the computer for 4 hours this afternoon, and during that time tpf's varied from 14:55 (a new low) and 25:50 (a new high). Since I made the first post in this thread I've been using the computer (though nothing using more than a fraction of one core), and TPFs have varied from 15:25 to 24:50.

 

[Kendrak]

There is something tumbling around eating up CPU cycles.

 

[Grandpa_01]

Um just curious but are you using the v7 PPD monitor to determine your TPF if you are don't it does not work correctly yet. And yes it varies widely use HFM to monitor your your client. Use musky's install guide to install HFM it works on 12.04 but HFM does have some problems on 12.04. When it starts up and the popup come up about having errors just click send error report then HFM will start in a minuet or two. You will have to manually refresh the client in HFM but it does work. http://hardforum.com/showthread.php?p=1037125479#post1037125479

 

[Quisarious]

I was using the control module to get those TPF's (from the log screen). I went back and checked the actual network traces and it is much more consistent.

Looking back over the last three 6903's, tpf's were 18:21 (the same average I computed using the control module log), but SD was only 27 seconds (as opposed to 2:40 using the module log) and does seem to be correlated with other uses as would be expected.

So it seems it is just a reporting issue in the V7 control module.

Thanks so much.