Julia HPC may need a kick(start)

You may or may not agree, but Julia is found lacking here:
Julia Still Not Grown Up Enough to Ride Exascale Train - The Next Platform

Both the blog post and the preprint actually seem quite a bit more positive than I would have expected given the title.


It is a bit tongue-in-cheek on my part, for sure.

One thing worth noting is that the performance issues in the paper come from GPU code where the authors explicitly say that the LLVM code looks good, so it’s unclear if anyone is capable of generating good code for this system other than just calling the AMD primitives.

1 Like

I agree. Note that the title of the preprint (“Julia as a unifying end-to-end workflow language
on the Frontier exascale system”) is much more positive than the one of the blog post.

Also, we’re talking exascale here. While it’s great that people are considering Julia for this, I’m already happy if we can play a good role in the “regular HPC” world.


I am very interested in the very short section on Jupyter notebook tests. I had expected a lot more given the introduction at the start of the article.
I think this statement tells us quite a lot about the potential of Julia
“This instance demonstrates Julia’s potential as a unified language that can seamlessly connect different stages of an HPC workflow, from computational simulation to data

I am going to go out on a limb here - traditional HPC workflows are import model / create a mesh / run the solver / run a post-processing program to produce results/pictures/movies

Then users either use Visualisation nodes remotely or indeed copy results files to storage which is accessible by Windows desktops/laptops.
If we can see that we can use the same ‘toolkit’ for visualisation then that is a great plus for Julia.

1 Like