Re: Linux stability

From: Bill Ross <>
Date: Fri 22 Jun 2001 13:51:17 -0700 (PDT)

        I am doing a 10 ns MD simulationon a 10bp oligonucleotide on a
        single pentium III 1000Mhz.
        I'm encountering trouble with RMS deviation too large, cannot
        reset coordinates
        FATAL error
        I understand the problem is that my molecules has a too strong
        deviation so the coordinate resetting cannot be done.
        But, i ran, from a previous restart file, the same calculation on
        my indigo2 (R10000), and it worked fine.

Is it repeatable on the PC?

        It could be a precision problem, but i didn't saw any warning
        about this kind of difference between x86 architecture and SGI's

Shouldn't be that..

        First of all, i would like to know if the problem i encounter
        could be due to the architecture.

If it's repeatable, it could be due to the architecture but in a
trivial way - trajectories diverge due to different floating point
implementations. It used to be that the only 2 architectures that
gave exactly equal results over a whole trajectory were HP & PC/g77,
so it would be interesting to see if it failed the same way on an HP.
Also it'd be interesting to diff the sgi/pc outputs & watch the

However, if it's repeatable, I think it more likely that there is
some bug at the compiler/library level on your PC. Possibly updating
versions would help.

        Is there a difference in sander_classic and the MA Young sander,
        from the center mass resetting point of view.

I'm not familiar with the MA Young version.

        Last point :
        Is it conceivable to finish an MD simulation which began under
        intel/linux (32-bit) on an indigo2/IRIX (i.e. on another platform

Bill Ross
Received on Fri Jun 22 2001 - 13:51:17 PDT
Custom Search