Re: AMBER: Implicit precision in sander vs architecture

From: Teletchéa Stéphane <>
Date: 17 Oct 2003 18:15:13 +0200

Le ven 17/10/2003 à 18:05, David A. Case a écrit :
> On Fri, Oct 17, 2003, Teletcha Stphane wrote:
> > I have recently seen in my dynamics simulations 'vlimit exceeded' on my
> > dual xeon although when using the same restart the calculation runs fine
> > on a power4 IBM.
> >
> > I know that implicit double precision is used in fortran code, but i
> > have no idea of what it means when the calculations are made on 32-bits
> > systems (pentium/athlon) or on 64-bits systems (power4, mips, ...).
> >
> There should be no differences (in either sander or pmemd) depending upon
> whether the *operating system* is 32 or 64-bit. In all cases, all floating
> point numbers are double precision.

I understand exactly this, but what precision does this provides
depending on the architecture : 10-10, 10-20, 10-4 ?

> You need to look someplace else to resolve your problem. A good first thing
> to try is to run short simulations on both machines (with the same restart
> file), setting ntpr=1, and look to see if you can identify any differences.

I'll see if i get time for this.

> Remember that long trajectories will never be the same on two different
> machines (due to the nature of Newtonian dynamics), so that it is certainly
> possible to have a vlimit problem on one machine but not another, even if
> there is nothing wrong on either machine. But repeated vlimit errors are
> certainly a cause for concern. Again, this has nothing to do with the sizes
> of addresses in the underlying OS.

It was only one error after 5.5ns without trouble, that's why i'm
wondering why. Taking the previous restart going on another machine (64
bits) made it run flawlessly. that's why i'm wondering where is the
cause ?

> ..good luck...dac



The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to

Received on Fri Oct 17 2003 - 17:53:02 PDT
Custom Search