AMBER Archive (2004)

Subject: RE: AMBER: AMBER 6 compilation problems!

From: Grace Li (w24li_at_student.cs.uwaterloo.ca)
Date: Fri Jul 30 2004 - 13:52:15 CDT

Hi Ross,

Thanks again for your explanation. I understand this aspect of differences
due to rounding differences, but do you think it would be ok to continue
sampling on my P4 cluster after already running 10 ns on the super
computer? Would this cause inconsistency in errors in my experiment?

Grace

On Fri, 30 Jul 2004, Ross Walker wrote:

> Dear Grace,
>
> > supercomputer. I noticed that after 900 NSTEPs the results starts to
> > differ as follows:
>
> Deviations in MD simulations between different architectures is perfectly
> normal. Especially with parallel runs. What happens is you get rounding
> differences that propagate over time and give you different (but equally as
> valid) trajectories. Typically in order to see if things are running
> correctly we only compare the first 10 or 20 steps of a simulation. The SGI
> architecture is also very different to a P4's architecture, one is 64 bit
> the other 32 bit for example, and so the rounding differences between the
> two machines will be more acute than if they were based on more similar
> hardware.
>
> Note, though, if you run the same simulation a 100 times on the SAME machine
> you should see identical trajectories. If you don't then something is very
> wrong.
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> | Department of Molecular Biology TPC15 |
> | The Scripps Research Institute |
> | Tel:- +1 858 784 8889 | EMail:- ross_at_rosswalker.co.uk |
> | http://www.rosswalker.co.uk/ | PGP Key available on request |
>
>
>
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu