• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

SMP@50% CPU Usage w/Project: 5101 (Run 0, Clone 137, Gen 113)

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

wingless

Member
Joined
Feb 5, 2008
Location
Houston, TX
Something ain't right. It took two days to finish one of these SMP WUs with low CPU usage, and it downloaded another early this morning! What the heck is going on? Is this WU not using 100% on purpose?

Code:
# Linux SMP Console Edition ###################################################
###############################################################################

                       Folding@Home Client Version 6.24beta

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: /home/johnnynismo/Desktop
Executable: ./fah6
Arguments: -smp -verbosity 9 

[13:07:10] - Ask before connecting: No
[13:07:10] - User name: texasoverclockers (Team 32)
[13:07:10] - User ID: B3E0ABF3DA7ED49
[13:07:10] - Machine ID: 1
[13:07:10] 
[13:07:10] Loaded queue successfully.
[13:07:10] 
[13:07:10] - Autosending finished units... [January 24 13:07:10 UTC]
[13:07:10] + Processing work unit
[13:07:10] Trying to send all finished work units
[13:07:10] Work type a1 not eligible for variable processors
[13:07:10] + No unsent completed units remaining.
[13:07:10] Core required: FahCore_a1.exe
[13:07:10] - Autosend completed
[13:07:10] Core found.
[13:07:10] Working on queue slot 00 [January 24 13:07:10 UTC]
[13:07:10] + Working ...
[13:07:10] - Calling './mpiexec -np 4 -host 127.0.0.1 ./FahCore_a1.exe -dir work/ -suffix 00 -checkpoint 15 -verbose -lifeline 5666 -version 624'

[13:07:10] 
[13:07:10] *------------------------------*
[13:07:10] Folding@Home Gromacs SMP Core
[13:07:10] Version 1.74 (November 27, 2006)
[13:07:10] 
[13:07:10] Preparing to commence simulation
[13:07:10] - Ensuring status. Please wait.
[13:07:27] - Looking at optimizations...
[13:07:27] - Working with standard loops on this execution.
[13:07:27] - Previous termination of core was improper.
[13:07:27] - Going to use standard loops.
[13:07:27] - Files status OK
[13:07:28] - Expanded 739037 -> 3855925 (decompressed 521.7 percent)
[13:07:28] 
[13:07:28] Project: 5101 (Run 0, Clone 137, Gen 113)
[13:07:28] 
[13:07:28] Entering M.D.
NNODES=4, MYRANK=0, HOSTNAME=linux-52mn
NNODES=4, MYRANK=1, HOSTNAME=linux-52mn
NNODES=4, MYRANK=2, HOSTNAME=linux-52mn
NNODES=4, MYRANK=3, HOSTNAME=linux-52mn
NODEID=3 argc=15
NODEID=2 argc=15
NODEID=1 argc=15
NODEID=0 argc=15
      Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
       Copyright (c) 1991-2000, University of Groningen, The Netherlands.
             Copyright (c) 2001-2004, The GROMACS development team,
            check out http://www.gromacs.org for more information.

        This inclusion of Gromacs code in the Folding@Home Core is under
        a special license (see http://folding.stanford.edu/gromacs.html)
         specially granted to Stanford by the copyright holders. If you
          are interested in using Gromacs, visit www.gromacs.org where
                you can download a free version of Gromacs under
         the terms of the GNU General Public License (GPL) as published
       by the Free Software Foundation; either version 2 of the License,
                     or (at your option) any later version.

([13:07:34] Calling FAH init
single precision)
[13:07:34] Read topology
starting mdrun 'CMF in water'
2000000 steps,   4000.0 ps.

[13:07:34] ocal files
[13:07:34] Completed 300000 out of 2000000 steps  (15 percent)
[13:07:34] Extra SSE boost OK.
[13:07:34] s
[13:07:34] Completed 300000 out of 2000000 steps  (15 percent)
[13:07:34] Extra SSE boost OK.
 
a1 core WUs are limited to 4 cores. The OS believes you have 8, not knowing the difference between a physical core and a logical one. It therefore reports you're using 50% of the processor. In reality you are using most of the 4 physical cores. THe a1 core probably averages about 80% cpu useage per core on 4 cores , unlike a2 which is about 99% of 4 or more cores.
 
Dammit I just figured out what the cause is. I had 1024MB allocated to my VMWare Player istead of 2048MB. I guess it downloaded the A1 thinking my system was weaksauce. I have 8GB of total system memory so I can have plenty of folding breathing room. I wouldn't have looked at those settings right now if I didn't read your post ChasR. Thanks for the info.
 
ChasR:
The OS believes you have 8
Why does it think there are 8 cores ChasR? We are both running Quad Phenoms, if Wing is folding on his sig rig, not i7's.

*EDIT* Ah, it's seeing the Linux install as a seperate system with 4 cores?



wingless:
Dammit I just figured out what the cause is


Not sure if that is the reason. I get the occasional a1 WU assigned to both my PC's. It seems random. No change in hardware or settings.

And yes, they are slow and annoying!
 
Last edited:
Actually I assumed he was using an i7 and didn't read his signature. So I was totally off on the matter, correct but for a different machine. It may well be that the a1 cores, p510x in particular isn't as efficient as I thought, at least on AMD rigs.

The a1s are hitting Linux rigs because the a2 servers have been on the fritz lately.

As for memory, the largest reported memory for a2 WUs is 600 MB, some are 200 and many are 64. Reporting 1000 qualifies you for assignment of every WU. Aside from what you enter in config (entering nothing defaults to all), the AS doesn't know how much memory you have really have.
Edit,
Looking at a C2D running a p5102, its using an average of 80% of both cores, but it will drop much lower for brief periods.
 
Last edited:
Actually I assumed he was using an i7 and didn't read his signature. So I was totally off on the matter, correct but for a different machine. It may well be that the a1 cores, p510x in particular isn't as efficient as I thought, at least on AMD rigs.

The a1s are hitting Linux rigs because the a2 servers have been on the fritz lately.

As for memory, the largest reported memory for a2 WUs is 600 MB, some are 200 and many are 64. Reporting 1000 qualifies you for assignment of every WU. Aside from what you enter in config (entering nothing defaults to all), the AS doesn't know how much memory you have really have.
Edit,
Looking at a C2D running a p5102, its using an average of 80% of both cores, but it will drop much lower for brief periods.

LOL I'm so glad you realized your mistake. My original post to you went something like this:
I quickly edited it and posted the equivalent of a polite smile and slowly backed away as you would do if you were talking to somebody that was insane.

ANYWAYS, It is interesting to see that the A1 cores are not efficient on AMD architecture. Frames are ~15 minutes per. I feel like I'm wasting my time and electricity on this core. Hope the SMP servers come up ASAP.
 
I pulled a number out of the air earlier. I'm actually at 23 minutes per frame:
Code:
[16:49:25] Folding@Home Gromacs SMP Core
[16:49:25] Version 1.74 (November 27, 2006)
[16:49:25] 
[16:49:25] Preparing to commence simulation
[16:49:25] - Ensuring status. Please wait.
[16:49:42] - Looking at optimizations...
[16:49:42] - Working with standard loops on this execution.
[16:49:42] Examination of work files indicates 8 consecutive improper terminations of core.
[16:49:43] - Expanded 739037 -> 3855925 (decompressed 521.7 percent)
[16:49:43] 
[16:49:43] Project: 5101 (Run 0, Clone 137, Gen 113)
[16:49:43] 
[16:49:43] Entering M.D.
NNODES=4, MYRANK=1, HOSTNAME=linux-52mn
NNODES=4, MYRANK=3, HOSTNAME=linux-52mn
NNODES=4, MYRANK=2, HOSTNAME=linux-52mn
NNODES=4, MYRANK=0, HOSTNAME=linux-52mn
NODEID=1 argc=15
NODEID=2 argc=15
NODEID=3 argc=15
NODEID=0 argc=15
      Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
       Copyright (c) 1991-2000, University of Groningen, The Netherlands.
             Copyright (c) 2001-2004, The GROMACS development team,
            check out http://www.gromacs.org for more information.

        This inclusion of Gromacs code in the Folding@Home Core is under
        a special license (see http://folding.stanford.edu/gromacs.html)
         specially granted to Stanford by the copyright holders. If you
          are interested in using Gromacs, visit www.gromacs.org where
                you can download a free version of Gromacs under
         the terms of the GNU General Public License (GPL) as published
       by the Free Software Foundation; either version 2 of the License,
                     or (at your option) any later version.

(single precision)
[16:49:49] Calling FAH init
[16:49:49] Read topology
[16:49:49] (Starting from checkpoint)
starting mdrun 'CMF in water'
2000000 steps,   4000.0 ps.

[16:49:49]  out of 2000000 steps  (15 percent)
[16:49:49] Extra SSE boost OK.
[16:49:49] s
[16:49:49] Completed 300000 out of 2000000 steps  (15 percent)
[16:49:49] Extra SSE boost OK.
[17:04:50] Timered checkpoint triggered.
[19:36:41] Writing local files
[19:36:41] Completed 320000 out of 2000000 steps  (16 percent)
[19:51:42] Timered checkpoint triggered.
[19:59:35] Writing local files
[19:59:35] Completed 340000 out of 2000000 steps  (17 percent)
 
Yea, I've pulled two of these in a row. My Q6600 gets 50% on them with about a 22min per frame. Really kills point production spending 2 days for ~ 2200 points.

There are a few other a1 projects that decimate production like this. Hopefully they'll get around to optimizing the packets to make full use of the processor cores.
 
Back