It has definitely changed since 2016 because of Nx. It depends on the type of work/algorithms you want to apply, but with Nx you can do numerical computation efficiently on individual nodes (especially if they also have a GPU) and thanks to OTP you can easily communicate between the nodes to coordinate the work. We did that to run some evolutionary algorithms using HPC, published in https://dl.acm.org/doi/10.1145/3512290.3528753 : )

Hi Patrick, the most recent effort that I am aware of is FLAME Slurm: GitHub - marcnnn/flame_slurm_backend: A FLAME Backend for Slurm.

I have expressed interest for running the Erlang distribution over RDMA, and if you are either a student or a supervisor interested in taking this forward, you can reach out to me.

Finally, we are preparing to implement GPU-to-GPU operations directly in Nx.

Which supercomputer did you run your HPC code on

Prometheus (Cracow, Poland) : )

Do you know of any other publications, articles or books that have content on Elixir and HPC?

Nothing in particular, especially related to Nx. I think there are a few papers that used Erlang and HPC, like this one, you can check out the author to see if there’s anything relevant for you.