Running model on multiple GPU cards

[wugaxp]

Thank you for the excellent model, which balances speed and accuracy!
I have a workstation with two NVIDIA GeForce RTX 2060 Super cards, each with 8 GB of memory. This is quite limited for general-purpose MLFFs. Is there a method to run the model on two (or more) GPGPU cards? I believe this feature would be quite useful for many users, as even with H100 cards, the accessible system size is still limited.

[yanghan234]

Thank you for raising this point. Unfortunately, MatterSim does not currently support running on multiple GPUs. We understand that this feature would significantly accelerate many simulations. However, we do not have the resources to prioritize this feature at the moment. In the meantime, we recommend running production simulations on high-performance devices like the V100 or A100 GPUs.

We warmly welcome contributions from the community. If you are willing to contribute to this feature, we would be very happy to support and collaborate with you.

[wugaxp]

Thank you for the reply. It's a bit disappointing to learn that the current model cannot run on multiple GPUs, as it would greatly help us extend the spatial and temporal scales of the system we're interested in. Many other good MLFFs also cannot run on multiple GPUs...