[phenixbb] Parallella

Nathaniel Echols nechols at lbl.gov
Sat Oct 27 11:27:03 PDT 2012

On Sat, Oct 27, 2012 at 7:30 AM, Morten Groftehauge
<mortengroftehauge.work at gmail.com> wrote:
> So the Parallella project seems like it is going forward (with the smaller
> processor first off)
> http://www.kickstarter.com/projects/adapteva/parallella-a-supercomputer-for-everyone
> I figure that phenix would have problems with low amount of RAM but are
> there other reasons that Phenix and Parallella might not play well together?

I think it will have the same problems as GPU acceleration of MX code.
 Most of the components in Phenix that benefit from parallel systems
work on a much coarser level (i.e. multiple CPU cores, or even
clusters), and most of these have already been parallelized (composite
omit map generation, building, ligand fitting, MR searches, Rosetta
jobs, refinement weight optimization, etc.).  Any further improvements
in speed will require improving the underlying algorithms (which is
definitely possible, if complicated), rather than adapting them to
specialized hardware.

There are scenarios where these parallel architectures could be
useful, and we've explored a few of them here.  But they're mostly
things that we simply wouldn't bother doing otherwise, not the kind of
stuff most crystallographers run every day.


More information about the phenixbb mailing list