6.24 only uses half of processor power?

[mattyf]

I just installed 6.24, and after running it for a while noticed that Activity Monitor reports CPU usage at only 50% or so. Looking at the log, it does seem to start only four nodes... I'm guessing it's only running on one of the two processors? I'm not sure what I need to do to get it to run on both. Any help would be appreciated...

I'm running it on a 2 x 3.2GHz Quad Core Intel Xeon, running 10.5.5, with 8 Gigs RAM.

Thanks!

[max]

running the

Code: Select all

-smp 8

flag?

A1 core WUs only use 4 cores. To utilize all cores you'll need to wait for an A2 core or run dual -smp 4 clients

[mattyf]

In the log it says:

Executable: ./fah6
Arguments: -smp 8 -local -verbosity 9

Is that right?

It also says: Calling './mpiexec -np 4 -host 127.0.0.1 ./FahCore_a2.exe -dir work/ -suffix 01 -checkpoint 15 -verbose -lifeline 852 -version 624'

[toTOW]

It looks like you got a WU designed for Quad cores ... the servers might be out of 8 cores WUs (or you added the -smp 8 flag after getting the current WU)

[mattyf]

could be both! I don't know what the hell I'm doing!

this WU finishes tonight, so I'll see what I find in the morning! Thanks!!!

[bruce]

In the log it says:

Executable: ./fah6
Arguments: -smp 8 -local -verbosity 9

Is that right?

It also says: Calling './mpiexec -np 4 -host 127.0.0.1 ./FahCore_a2.exe -dir work/ -suffix 01 -checkpoint 15 -verbose -lifeline 852 -version 624'

Please describe the hardware. Specifically, is this running on a CPU with 8 cores?

[MtM]

In the log it says:

Executable: ./fah6
Arguments: -smp 8 -local -verbosity 9

Is that right?

It also says: Calling './mpiexec -np 4 -host 127.0.0.1 ./FahCore_a2.exe -dir work/ -suffix 01 -checkpoint 15 -verbose -lifeline 852 -version 624'

Please describe the hardware. Specifically, is this running on a CPU with 8 cores?

It's given in the opening post, 2 xeon quad's = 8 cores.

toTow I been told that a2 would scale to use all cores on any setup, so he must have edited it in after this wu was downloaded and started. If the wu is started it remember the amount of cores used and will continue to use them for the remainder of the work unit, regarding if you got a skulltrail for instance and plug in another quad halfway through it. So I think there are no x cores a2 wu's, they will all run on any configuration.

I think you where in that thread as well, I asked how a2 would handle the 6 core dunnington's coming this year.

[mattyf]

ok ok I admit it. I changed it to smp -8 and relaunched it after it started the first WU. I'm a newbie! I'm just trying to help cure cancer, is that so wrong?

So now overnight it downloaded a new work unit. This one it reports as FahCore_a1. It's using 4 cores. Activity Monitor reports 1 core running at 99%, the other three are running at around 66%.

[MtM]

It's not wrong, and I'm happy to have you onboard, just thought toTow's suggestion is impossible since afaik a2 wu's should scale to all cores ( if you started them with -smp 8 in your case ).

Sorry to hear you got an a1 wu, but those need folding as well. It does mean that with alternating a2 and a1 wu's being assigned you're not being as productive as you could be with a steady supply of either. Running one a2 wu or two a1 at the same time both are nice options, but not possible with the current state of assignment servers/workserver/available projects

[mattyf]

Cool thanks. I'll wait and see what happens the next time I get an a2 then. Appreciate the help!

[mattyf]

Well, in the time since the last post, I've gotten an additional 4 WUs and they have all been a1s... Is it possible that I somehow installed it incorrectly? Or are a2 units relatively rare? Should I uninstall and reinstall not messing up the original params this time?

Also, for my own curiosity, why do the a1 units not use the full power of the four cores they are running on?

[toTOW]

There's not a lot of A2 WUs at the moment ... that's why you don't see them ...

I don't know why (technically speaking) A1 WU doesn't use the full power of the 4 cores ... A1 was build on an older version of Gromacs, and my first observations indicates that the core seems very dependent to communications through MPI interfaces. This core likes high speed inter-core communications and fast memory accesses.

[mattyf]

So I finally got dealt an a2 core, and the same thing happened... it's using only 4 of the 8 cores... about half the mac's processing power according to Activity Monitor.

This is the way I start FAH:
./fah6 -smp 8 -local -verbosity 9

Here's the startup log:
[14:26:14] + Processing work unit
[14:26:14] At least 4 processors must be requested.Core required: FahCore_a2.exe
[14:26:14] Core found.
[14:26:14] - Using generic ./mpiexec
[14:26:14] Working on queue slot 08 [March 24 14:26:14 UTC]
[14:26:14] + Working ...
[14:26:14] - Calling './mpiexec -np 4 -host 127.0.0.1 ./FahCore_a2.exe -dir work/ -suffix 08 -checkpoint 15 -verbose -lifeline 346 -version 624'

[14:26:14]
[14:26:14] *------------------------------*
[14:26:14] Folding@Home Gromacs SMP Core
[14:26:14] Version 2.04 (Wed Jan 21 09:16:08 PST 2009)
[14:26:14]
[14:26:14] Preparing to commence simulation
[14:26:14] - Looking at optimizations...
[14:26:14] - Working wius OK
[14:26:14] ndard loops on this execution.
[14:26:14] - Files status OK
[14:26:15] - Expanded Called DecompressByteArray: compressed_data_size=Called DecompressByteArray: compressed_data_size=4845590 data_size=24003985, decompressed_data_size=24003985 diff=0
[14:26:16] - Digital signature verified
[14:26:16]
[14:26:16] Project: 2672 (Run 0, Clone 173, Gen 90)
[14:26:16]
[14:26:16] Entering M.D.
NNODES=4, MYRANK=3, HOSTNAME=zMatt.local
NNODES=4, MYRANK=0, HOSTNAME=zMatt.local
NNODES=4, MYRANK=1, HOSTNAME=zMatt.local
NNODES=4, MYRANK=2, HOSTNAME=zMatt.local
NODEID=0 argc=20
NODEID=1 argc=20
NODEID=3 argc=20
NODEID=2 argc=20
G R O M A C S (-:

Groningen Machine for Chemical Simulation

VERSION 4.0.3_pre (-:


Written by David van der Spoel, Erik Lindahl, Berk Hess, and others.
Copyright (c) 1991-2000, University of Groningen, The Netherlands.
Copyright (c) 2001-2008, The GROMACS development team,
check out http://www.gromacs.org for more information.


mdrun (-:

Reading file work/wudata_08.tpr, VERSION 3.3.99_development_20070618 (single precision)
Note: tpx file_version 48, software version 58

NOTE: The tpr file used for this simulation is in an old format, for less memory usage and possibly more performance create a new tpr file with an up to date version of grompp

Making 1D domain decomposition 1 x 1 x 4
starting mdrun '22890 system'
22750000 steps, 45500.0 ps (continuing from step 22500000, 45000.0 ps).
[14:32:03] 2500 out of 250000 steps (1%)



Any thoughts?

[susato]

If you use the -smp flag without a number, it assumes 4 cores. That may be the case you are seeing now. If you have the -smp 8 flag set then it's odd that the client is only using 4 cores. In that case I'm not sure what is going on - but until it's sorted out, you may wish to run two instances of the SMP console client set to 4 cores each. (be sure to set each to a different machine ID and place the two instances in different folders) That will max out your 8-way Mac Pro no matter whether you are getting core_a1 or core_a2 units.

[mattyf]

I'm definitely using the -smp 8 flag. However, the very first time I ran it I left that flag out. I wonder if its possible that that inadvertently set some sort of persistent preference for four cores?