SLURM config for multiple nodes

[DavidHoover]

Does anyone have an example of how to configure CryoSPARC running under SLURM and allocating multiple nodes for a single job? I am not clear how CryoSPARC distributes work across multiple nodes. Does it use MPI, ssh or something unique? Is there a limit to --ntasks-per-node for GPU nodes?

Thanks,
David

[leetleyang]

AFAIK, there is no multi-node support. See below.

Recommendation for lane configuration

@sbliven, all cryoSPARC jobs are python processes that, depending on the job type, use multiple threads or create multiple python subprocesses. We do not use openMP or MPI. There is currently no support for multi-node processing, only multi-GPU multiprocessing on the same node. All concurrency is done right now by us in the code itself (i.e. the spawning of threads or multiprocessing workers) depending on the parameters set (there is usually a parameter controlling the number of GPUs to paralle…