[gmx-users] mixed single-precision GPU and double-precision CPU calculation

Berk Hess gmx3 at hotmail.com
Fri Nov 30 11:56:42 CET 2012


As the new Verlet cut-off scheme, also used on the GPU, uses separate non-bonded
coordinate and force buffers, implementing this will require very little code changes.
All real variables in the nbnxn_atomdata_t data structure and in all code operating
on them should be replaced by floats. We might implement this, but currently we are
far too busy with other things.



> From: makoto-yoneya at aist.go.jp
> To: gmx-users at gromacs.org
> Date: Thu, 29 Nov 2012 09:37:41 +0900
> Subject: [gmx-users] mixed single-precision GPU and double-precision CPU calculation
> Dear GROMACS peoples:
> I know the current gromacs-4.6 can't build with both GPU and double
> precision being enable.
> I'm wondering is it possible to modify mixed single-precision
> (non-bonded) GPU calculation and double-precision CPU calculation?
> I suppose this combination would be practically useful when we needs high
> precision constraint
> calculation (with SHAKE or LINCS etc.).
> If it is possible by modifying some routine, please teach me the routine to
> modify.
> I would like to try it.
> Thank you for advance.
> Makoto Yoneya, Dr.
> AIST, Japan
> Makoto Yoneya, Dr.
> http://staff.aist.go.jp/makoto-yoneya/
> --
> gmx-users mailing list gmx-users at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-users
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
> * Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists

More information about the gromacs.org_gmx-users mailing list