[gmx-users] Gromacs on Intel Xeon and AMD Opteron

[Xu, Jianqing]

Thanks for the comments, Mark.

Sorry that I did not explain my question clear enough.

I am actually talking about post-equilibration time average. I understand that some instantaneous deviations are expected, but the simulations should be correct as long as the ensemble averages are the same (or close enough), but the differences that I observed seemed beyond that.

I firstly realized the total energy reported from two servers are obviously different (again, two simulations started from the same *.tpr file, and the executable on each server were compiled independently using the same arguments). I then checked the potential energy and kinetic energy as shown the below link:

https://docs.google.com/file/d/0B7MGQ6N_2Ln7emRySENsZU40d2s/edit?usp=sharing


Looks like the differences are from the potential energy. I then took a quick look at the LJ (SR) and Coulomb (SR) components, and it looks like the Coulomb (SR) component has big deviation which beyond the numerical errors. There are other differences, but looks like the Coulomb (SR) is the most obvious one. I was not able to understand how this can happen.

https://docs.google.com/file/d/0B7MGQ6N_2Ln7bzB4YVhvMjdsNkE/edit?usp=sharing

I hope the figures can clarify my situation better, and I will appreciate if you or anyone else on the list could provide more suggestions.

Thanks a lot!

Regards,

Jianqing



________________________________________
From: gmx-users-bounces at gromacs.org [gmx-users-bounces at gromacs.org] on behalf of Mark Abraham [mark.j.abraham at gmail.com]
Sent: 11 September 2013 10:56
To: Discussion list for GROMACS users
Subject: Re: [gmx-users] Gromacs on Intel Xeon and AMD Opteron

It's not clear whether you are reporting single points or
post-equilibration time averages, but either way you must expect
differences. MD simulations are chaotic. However, the long-time
ensemble averages should agree well - that's the point of the
simulation.

Mark

On Wed, Sep 11, 2013 at 10:35 AM, Xu, Jianqing <xuj at medimmune.com> wrote:
> Dear all,
>
> This is my first time here. I apologize if I am not aware any rules for posting a new message.
>
> I was testing Gromacs on two servers, one server has AMD Opteron processor (Server 1), and the other has Intel Xeon processor (Server 2). The simulation system that I tried is the lysozyme system from the online tutorial (http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin/gmx-tutorials/lysozyme/).
> Briefly, it has a lysozyme in 10824 water, and 8 Cl-. Box size is around 6.99 nm.
>
> I finished all the preparation steps advised by the tutorials, and tested the final NPT production runs on two servers, using the same final input file "md_0_1.tpr". It is strange that the results from Server 1 (AMD) has higher potential energy than the results from Server2 (Intel), specifically, results from Server1 (AMD) has higher "Coulomb-(SR)" than Server2 (Intel) (  ~ -5.5*10^5  .vs.  ~ -5.7*10^5  kj/mol ).
>
> I was wondering if anyone could give me any suggestions about what could be the reasons causing this discrepancy.
>
> Some additional information:
> 1). Jobs run on my local computer, which also has Intel CPUs (Intel(R) Xeon(R) CPU E5620  @ 2.40GHz), reported the same results as the Server2 (Intel(R) Xeon(R) CPU X7550  @ 2.00GHz).
> 2). All installations (Gromacs and OpenMPI) on Server 1 and 2 were compiled independently, but with the same arguments.
>
>
> I understand that the information I provided may not be sufficient. If I should provide more specific information, please let me know.
>
> Any help will be highly appreciated!
>
> Thanks,
>
> Jianqing
> To the extent this electronic communication or any of its attachments contain information that is not in the public domain, such information is considered by MedImmune to be confidential and proprietary. This communication is expected to be read and/or used only by the individual(s) for whom it is intended. If you have received this electronic communication in error, please reply to the sender advising of the error in transmission and delete the original message and any accompanying documents from your system immediately, without copying, reviewing or otherwise using them for any purpose. Thank you for your cooperation.
> --
> gmx-users mailing list    gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
--
gmx-users mailing list    gmx-users at gromacs.org
http://lists.gromacs.org/mailman/listinfo/gmx-users
* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
* Please don't post (un)subscribe requests to the list. Use the
www interface or send it to gmx-users-request at gromacs.org.
* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
To the extent this electronic communication or any of its attachments contain information that is not in the public domain, such information is considered by MedImmune to be confidential and proprietary. This communication is expected to be read and/or used only by the individual(s) for whom it is intended. If you have received this electronic communication in error, please reply to the sender advising of the error in transmission and delete the original message and any accompanying documents from your system immediately, without copying, reviewing or otherwise using them for any purpose. Thank you for your cooperation.