GPU usage in FEP calculation

Thanks for your help pszilard

By the way, you seem to not be running PME on the GPU (-pme gpu), it seems it should be compatible with your settings, is it not? I suggest to try that since most of your CPU runtime is actually in PME not the CPU integration!

Cheers,
Szilárd

The -pme gpu doesn’t work in my settings since pme-order is 6 in my mdp file, so i get error message of PME GPU does not support: Interpolation orders other than 4

Do you need to use PME order 6?

Since we are running absolute FEP for a charged system, so we have kept pme order 6.

I don’t see why either the net charge och absolute free energy would require different PME settings. Have you observed significant errors with normal settings or do you have a reference?

We haven’t tested under normal settings and the reference article doi is 10.1021/jacs.6b11467

I see. I don’t see a need for such high PME accuracy though.

The best thing is to test yourself if you see any differences. Checking dH/dlambda at one or two lambda points should be sufficient.

1 Like

Thanks Hess

Will try to check it by changing the pme order.