GPU package using QUIP potential

Dear all,

I’m trying to use the GPU with a QUIP (GAP) potential. I know that the GPU package can’t use the Newton flag on, and that the QUIP potential demands Newton flag on. By saying that, I’m almost certain that I can’t use the GPU in this case. However, I need so bad to gain more computational power that I’m here to confirm that there’s really no way to make this works. Using the 29 Oct 2020 lammps version.

Best regards,

a) QUIP does not support GPU acceleration.
b) there is currently no clean way to work around the newton on/off issue when mixing GPU and non-GPU pair styles.

sorry for the bad news,

Dear akohlmey

Is it possible to utilize GPU acceleration for computing GAP potential in QUIP?


I am confused. You are replying to the post that already provides the answer to your question. Why ask again?

Dear akohlmey
Sorry for the confusion my reply caused. I meant to say that the QUIP package has been in development for three years since 2021, and I am looking forward to it being updated to be compatible with GPU acceleration.


There is next to nothing that can be done about it from the LAMMPS developers.

The ML-QUIP package is a very lightweight wrapper around libQUIP. Apart from some adjustments to internal changes in LAMMPS and some bugfixes in the C++ code, there has been no development of the ML-QUIP C++ code since 2015. The only changes were in the QUIP library itself, which we don’t control.

In general, given the way how LAMMPS is documented, any GPU support would be mentioned in the documentation. If you don’t see it, it isn’t there.

Thank you for your response.