i7-3820 Extremely Slow SMP

Moderators: Site Moderators, FAHC Science Team

i7-3820 Extremely Slow SMP

Postby Hikyuuri » Fri Sep 21, 2012 6:57 am

Hello.

I recently upgraded from a Core 2 Quad 9550 (2.83 GHz quad core) to a i7-3820 (3.6GHZ quad core w/ HT). I expected my PPD to skyrocket, but that is not the case.

Work units that I know I folded in less than a day on my old 2.83GHz quad core are taking WAY longer (upwards of 12 days). I used to get around 3000-6000 PPD (depending on what else I was doing on the machine), now all I get is 400-600PPD.

Right now, it is working on PRCG 7647 (48, 0, 28), with a base credit of 2250. It is giving me an estimated TPF of just over 3 hours.

I used to get WUs close to this point range (2000-2400) on my old CPU, and folded them in under 2 days (don't have exact numbers, never really looked that closely). Now, I have a faster processor, but it is uber slow. The settings are exactly the same, except that I now use SMP:8 instead of SMP:4 (due to the HT). I have max-packet-size set to big, and client-type set to bigadv (same as with the Q9550).

Hell, my i5 dual core laptop is folding WAY faster than my quad core desktop, and my laptop is only a 2.5GHz dual core. Right now, my laptop is running project 7808 (base credit of 1698), with a TPF of 22 minutes. It is also running Bigadv and big WUs (same as desktop), but for some reason, it is folding much faster (almost 7X the PPD).

Any ideas on why this is? I would like to keep folding, but right now, it is maxing out my CPU, but not giving me the points it used to.
Hikyuuri
 
Posts: 5
Joined: Tue Dec 13, 2011 6:01 am

Re: i7-3820 Extremely Slow SMP

Postby Zagen30 » Fri Sep 21, 2012 7:22 am

Yes, that is extremely slow.

How long have you been folding since the upgrade; is that 12 days quote what you've actually observed, or is it just how long the client is estimating it will take? If you look at the log files, does the TPF reasonably match the estimated TPF? There are still some known bugs with the TPF/PPD estimations that can cause them to be very wrong even while things are really running fine.

Have you check Task Manager to see if there's anything else using CPU time? SMP processing can be seriously impeded if even one thread is slower than the others, as the other threads need to wait for the slow thread to finish and sync up with it every so often. Since FAH runs at the lowest priority, many other tasks, even using a small but constant amount of CPU time, can cause frame times to skyrocket.

In a very similar vein, you didn't happen to add an AMD GPU during this upgrade, did you? They use a lot of CPU time when folding.

Posting the contents of the log file, including the ***System*** section near the top, in Code tags would help diagnose the problem.

Also, you can remove the bigadv argument. At this point in time bigadv is restricted to machines with 16 or more reported cores. The servers won't send you any bigadv work, but having the argument there isn't gaining you anything. You could change it to client-type:advanced, which would signal that you're willing to fold WUs that might be slightly less stable than those released to everyone, but you don't get any extra points or anything for doing so.
Image
Zagen30
 
Posts: 823
Joined: Tue Mar 25, 2008 1:45 am


Return to V7.1.52 Windows/Linux

Who is online

Users browsing this forum: No registered users and 2 guests

cron