Running on multiple GPU's?

[RHoejland]

Hey ANI devs,

I'm trying to run a somewhat large system, and run into memory issues. I have access to multiple A100 (80GB) GPU's on the same node, is it possible to run the model across multiple GPU's effectively increasing the available memory? I've read some comments about updates which would downscale the memory issues significantly, but i some date back to summer last year, is there any update on that?

I'm using PBC, and the last attempt (using a mixed system with amber-tip3p as the solvation forcefield. Currently this system tries to allocate ~420 GB onto the GPU, but i can probably reduce the system size enough that it'll fit onto 4x80GB.

/Rune