[gmx-users] Re: Secondary structure loss in implicit solvent simulations

[Michael Shirts]

> Do you get the same effects if you run a "normal" simulation on CPU and not
> GPU?  That information would be critical for properly diagnosing what's
> going on. If it's not GPU-specific, in all likelihood whatever you're doing
> is incorrect somewhere along the way.

Folllowing up on this -- if it's not the same trajectory as the CPU to
within single precision, then something is wrong.

So the energies should be the same to 5 digits for the first time
steps, and then gradually drift away. If it's not the same for the
first 5-10 steps, then there is something going wrong with the GPU
code or the setup for the GPU code.