Single vs. Double precision/forcefields

[Joe Leonard <jle $#at#$ theworld.com>]

 Folks,
 Has anybody done, or does anybody have pointers to a study
 on the effect of precision on forcefield energies and/or gradients?
 This would be most interesting on Intel and/or AMD chipsets
 (which seem to prefer single-precision up through P-III's)...
 Note to language bigots, please replace single/double precision
 with float/doubles as necessary :-).
 Happy new year to all!
 Joe
 P.S. I would also be interested in learning of any groups doing
 work on internal data representations of chemical entites, and
 how this would relate to how codes are written in the future...