[gmx-users] one molecule on more CPU's

[Erik Lindahl]

On Friday, November 15, 2002, at 07:53 AM, Anton Feenstra wrote:

> Hi All,
>
>
> I'm still trying to get my scaling up for the protein in water system.
> Now I (think I) remember Erik mentioning that it would be possible
> for a single molecule (e.g. a protein) to cross processor boundaries.
> Is this already in one of the distributed versions or is this still
> only CVS (or is it only yet in the planning)?
>

It's no problem for a molecule to cross a processor boundary - but the
shake blocks (we call them shake blocks even when we use lincs)
cannot cross CPU boundaries - yet.

It is actually possible to parallelize LINCS, and we will probably do
it when the rest of the better scaling is working. For now you can do
the same thing most other programs do: only constrain h-bonds.

PS: I wouldn't recommend the HEAD branch of CVS the next couple
of weeks, we're working rather hard on it...


Cheers,

Erik