How to run TWO GPU calculation for DEM-Engine

[Hiroaki Amakawa]

I would like to run a DEM simulation with GPU parallelization.

According to the documentation (<https://api.projectchrono.org/deme_usage.html>), it seems that simulations can be executed using two GPUs. Is there a way to explicitly specify and enable TWO-GPU execution?

I tested the simulation on a Linux PC cluster equipped with two GPUs. However, when monitoring the system during execution, it appeared that only one GPU was being utilized.

Could you please let me know if any additional settings, options, or configuration steps are required to run DEM simulations in parallel on two GPUs?

[Ruochun Zhang]

Hi Hiroaki,

One problem with the current DEME is that it will automatically use available GPUs and the user doesn't control it. Such behavior will be changed a bit later, but for now, if it sees 2 GPUs, it will use both of them.

So perhaps you should first check if nvidia-smi returns that both GPUs are visible. If they are, then whether the solver is using 2 GPUs can be seen from a solver initialization message "Number of total active devices". If it shows 2 there, it is using 2 GPUs.

Perhaps by monitoring, you referred to the power consumption metric. The two GPUs' streams are highly asynchronized and maybe in some systems you don't see them being busy at the same level. You could instead monitor if your simulation does run faster to determine if it is taking effect. On consumer GPUs, a duo should help; not necessarily so for stronger data center GPUs.

In any case, using two GPUs is certainly a way to scale your simulation a bit larger than one GPU can do, but whether it saves you time depends. Let me know if you observed something I did not explain.

Thank you!

Ruochun