Hi there,
>From what I've read and known, here in the list as well, one of the main
reasons why sander/pmemd demands double precision is SHAKE (at least from
the pmemd.cuda point of view).
Reading Amber11 manual and I verified that Amber uses only SHAKE, no SETTLE,
no LINCS, nothing else.
I saw the discussions here about double/single precision or even hybrid
precision (just reading about Anton, from Shaw's group).
When I started in MD, developing our own MD software, all was done in double
precision, then came Gromacs blowing up this paradigm. (I am aware that even
in Gromacs, there are routines that really requires double precision, e.g.
normal mode analysis).
I am starting this discussion here but I post in Gromacs list too, but
essentially I would like to understand better this double x single approach
in MD re accuracy. And of course correct what I said in the 1st paragraph
since I may be wrong in my assertion.
Thanks,
Alan
--
Alan Wilter S. da Silva, D.Sc. - CCPN Research Associate
Department of Biochemistry, University of Cambridge.
80 Tennis Court Road, Cambridge CB2 1GA, UK.
>>http://www.bio.cam.ac.uk/~awd28<<
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu May 13 2010 - 08:30:04 PDT
