This article in Nature Reviews talks about:
the challenges of adapting quantum chemistry software to these new architectures, including the fragmentation of the software stack, the need for more efficient algorithms (including reduced precision versions) tailored for GPUs, and the importance of developing standardized libraries and programming models.
does anyone have the pdf?
They seem to not be citing the paper that just won the Gordon Bell prize that did ab initio molecular dynamics using MP2 potentials and achieved an exaflop of FP64 on Frontier.
I found a preprint of the paper and it seems to have quite different references, if someone can share the pdf that would be amazing, I am keen on reading this since this is what I do.
Yeah further more, this is a topic that is near to my heart since I have both contributed to a Fortran and C++ based application. Fortran suffered greatly due to the lack of GPU support in the early days of the Department of Energy’s Exascale Computing Project…still connecting to cuda and hip via Fortran is a royal pain in the neck.
I suppose it is not legal to share the Nature paper…
However, the preprint below does mention the Gordon Bell paper you are referring to.
Interesting read!! Is curious that they didn’t analyze the GNU tool chain, but only the vendors intel, AMD and Nvidia.
Another related article in the same vane: Evaluating GPU Programming Models for the LUMI Supercomputer | SpringerLink
For reference, the arXiv version seems to be here:
and the HTML version:
Some of QMC related codes (mentioned in the paper) may be here:
and the 2024 Gordon Bell prize page (which mentions molecular fragmentation + MP2).
ah! my library had an earlier version that did not sorry about that
