Thanks Jason for your background on this issue. It looks like I really
have to experiment with -np 2, 4 etc.
Bob Hopkins
-----Original Message-----
From: Jason Swails [mailto:jason.swails.gmail.com]
Sent: Tuesday, April 12, 2011 12:00 PM
To: AMBER Mailing List
Subject: [Possible Spam: 10%] Re: [AMBER] Processors, cores and threads
Hi Bob,
You can actually run as many threads as you want, the hyperthreading
label
just says that the cores can handle 2 threads fairly efficiently. How
efficiently, I'm not sure.
As far as actual timing is concerned, you'll see competing effects.
First,
Amber programs won't scale perfectly. Even with 8 physical cores you'll
almost never see 8x improvement. Then you have the effect that 4
physical
cores with 'hyperthreading' will likely be less efficient than 8
physical
cores. I'm not sure what the *actual* difference is. (The only way to
really nail that down would probably be running 4 threads; once forcing
2 of
the cores each to run 2 threads, and the other one to use all 4 cores.)
That said, your simulations will probably finish faster running 8
threads
than they will running 4, but not as much faster as it normally should.
>From another angle, with the hyperthreading, I can imagine you'd be
fine
running Amber with 4 threads while using your computer for other things.
Of course, this is all speculation (hopefully intelligent speculation
;)).
The only way to substantiate these claims would be to actually run the
benchmarks and just be careful not to pollute them with other processes
stealing computing resources. I would actually be interested to see how
well the hyperthreading performs (but that would require controlling
mpirun
carefully to make sure that threads are not distributed 'optimally').
Hope this helps,
Jason
On Tue, Apr 12, 2011 at 9:14 AM, Hopkins, Robert <hopkins.uhcl.edu>
wrote:
> Amber users and developers,
>
>
>
> I have just installed Amber11 on my new MacBook Pro (I finally
switched
> from the Windows environment and am completely new to Mac). This
> machine has 1 processor (Intel i7) with 4 cores and, through the
> implementation of hyperthreading, can run 8 threads. My question is
> regarding the option -np (number of processors?) for parallel runs:
> what is the actual/optimal number that should be used for this value?
I
> would greatly appreciate any comments or suggestions.
>
>
>
> Thanks!
>
> Bob Hopkins
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Apr 12 2011 - 11:00:03 PDT
