[gmx-users] Free energy on GPU?

[Олег Титов]

Dear all.

I'm trying to estimate relative solvation enegry of brombenzene (relative
to benzene) via thermodynamic integration. Can I perform such calculation
in GROMACS with GPU acceleration?

As I currently understand to make use of GPU one need to use Verlet cutoff
scheme. When I do so with pme electrostatics everything works fine until
free energy calculation stage. Something strange happens there: the
temperature of the system increases with increase of lambda value with no
respect to thermostat (300 at lambda = 0.0 -> 500 at lambda = 1.0).
Interesting fact that with reaction-field electrostatics everything seemed
to work fine but now I cannot be sure that these results are reliable.

As I understood from manual there is some issue with Verlet cutoff scheme
and free energy calculations. Can I perform TI calculation in GROMACS with
GPU acceleration or should I use group cutoffs and forget about GPU?

Thanks a lot,
Oleg Titov