AMBER Archive (2005)

Subject: Re: AMBER: sander error

From: Robert Duke (rduke_at_email.unc.edu)
Date: Tue Oct 18 2005 - 09:49:05 CDT

Okay, guys, this is a really really huge system, bigger than supported by
either the prmtop or inpcrd file formats I believe, so even pmemd is not
going to rescue you. Both prmtop and inpcrd have an i6 format for the
number of atoms, so this means we can handle up to 999,999 atoms. PMEMD 8
should handle this large (999,999 atoms) a system I believe, even with a 1GB
single processor machine, but you would have to compile a version with
pairlist compression enabled, which means that you DON'T have
DIRFRC_BIGCACHE_OPT defined. However, this will be a really slow simulation
on a single processor, due to its shear size, so it should really be run on
4 or more processors, and then you should be able to choose the default
configuration, whatever it is. In the amber 9 codebase in development, I
actually no longer support pairlist compression because the computational
overhead is higher than the caching benefits on machines with 500 KB or
bigger caches in general, and really these huge systems are just too slow to
run on single processors (in other words, you really need multiple
processors, and then you don't need pairlist compression). So knock this
thing down to < 1,000,000. As Ross says, realistically folks are reaching
into the 100,000-200,000 atom range but not much higher due to the run
times. I have run pmemd on up to 500,000 atoms myself on a single
processor, but have not gone higher.
Regards - Bob Duke

----- Original Message -----
From: "Ross Walker" <ross_at_rosswalker.co.uk>
To: <amber_at_scripps.edu>
Sent: Tuesday, October 18, 2005 10:23 AM
Subject: RE: AMBER: sander error

> Hi Yam
>
>> The system contains about 1.4M atoms with octahedron
>> water and ions. And the attempting minimization is
>> only step no. 1, therefore setting ntx=1 (default)
>> (not even on MD runs). I have checked the last line of
>> my inpcrd file and is clear that the water box is
>> octahedron.
>> Anyone has any idea on what is going on? could it be
>> the limited number of atoms in sander? Please help.
>
> This is a huge system, read at least an order of magnitude larger than
> what
> we would consider a very large system. I don't think sander has even been
> tested with a system this large. The errors you are getting are almost
> certainly due to the system size, possibly a memory error due to some
> routine not expecting such a large system and not having a big enough
> buffer. E.g. reading of the prmtop file or inpcrd file etc...
>
> You could try pmemd, this is designed to handle large periodic systems but
> 1.4 million atoms may be larger than this was ever tested for. Note, with
> this sort of size system you are going to need some serious computing
> power
> (> 128 cpus of a SMP machine) to have a chance of doing simulations long
> enough to get meaningful results. Even then you will burn a LOT of cpu
> time.
>
> I suggest you try looking at a subset of the system and just run some
> simulations on that and see if you can confer any macroscopic information
> from this.
>
> 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 |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>
> -----------------------------------------------------------------------
> 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