AMBER Archive (2008)

Subject: Re: AMBER: massively parallel computation

From: Robert Duke (rduke_at_email.unc.edu)
Date: Tue May 20 2008 - 09:32:22 CDT

Hi Adrian,
Thanks for the support :-) I think a pointer to amber.scripps.edu, see the
benchmarks lists, should suffice. I am tied up doing other hopefully more
career-enhancing stuff at the moment than the 10 benchmarks, but I would
remark that 10 is typically between 10 and 50% faster than 9 depending on
your hardware, your ensemble (with NVE performance > NVT performance > NPT
performance), and other details of your problem. The comparisons have to be
made of pmemd 9 running today on hardware vs pmemd 10 running today on
hardware. I make this comment because in a number of instances, hardware
"out there" has been "upgraded" and the impact has been to make things
slower. So, IBM apparently dinked with the OS on the sp5's and now they are
something like 5% slower than they used to be (probably some compromises to
better support threading, but I am not sure). Cray waffled in their
commitment to 1 node = 1 cpu and started putting multiple cpu's in 1 node
(so the xt3 now has been "upgraded" to dual cpu's in a node) and it is
somewhere around 5% slower. Various intel commodity clusters have been
"upgraded" to multicore, going from dual cpu's running at 3.2-3.6 GHz to 4-8
running in the low 2 GHz range - a real loser, but what you can do to
counteract this little disaster is only use 6 out of the 8 cpu's. And so
on.
Regards - Bob

----- Original Message -----
From: "Adrian Roitberg" <roitberg_at_qtp.ufl.edu>
To: <amber_at_scripps.edu>
Sent: Tuesday, May 20, 2008 10:11 AM
Subject: Re: AMBER: massively parallel computation

> Oh man, you got yourself in trouble now ...
> Amber and more particularly PMEMD already use hundreds of cpus VERY
> efficiently !
>
> It might be time to post to this list the benchmarks showing how good
> pmemd really is against the other supposedly faster programs.
>
> Ross ?
> Bob ?
>
> Cheers
>
> Adrian
>
>
> Mingfeng Yang wrote:
>> Recently, a few algorithms have been developed to enable massively
>> parallel
>> computation which can efficiently use hundreds of CPUs simultaneously for
>> MD
>> simulation. For example, J Comput Chem 26: 1318–1328, 2005.
>>
>> Is there a plan to implement such algorithm in Amber/PMEMD? As computer
>> cluster is getting cheaper and cheaper, the cluster size keeps expanding
>> quickly as well. Such algorithms should be very helpful and indispensable
>> to
>> reach >ms scale simulation.
>>
>> Thanks,
>> Mingfeng
>>
>
> --
> Dr. Adrian E. Roitberg
> Associate Professor
> Quantum Theory Project and Department of Chemistry
>
> University of Florida PHONE 352 392-6972
> P.O. Box 118435 FAX 352 392-8722
> Gainesville, FL 32611-8435 Email adrian_at_qtp.ufl.edu
> ============================================================================
>
> To announce that there must be no criticism of the president,
> or that we are to stand by the president right or wrong,
> is not only unpatriotic and servile, but is morally treasonable
> to the American public."
> -- Theodore Roosevelt
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
> to majordomo_at_scripps.edu
>

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" (in the *body* of the email)
      to majordomo_at_scripps.edu