[QE-developers] Quantum ESPRESSO readiness for heterogeneous HPC?

[Paolo Giannozzi]

Hi
1. yes, but only a subset of the distribution can be considered 'ready':
see the QE-gpu project, https://gitlab.com/QEF/q-e-gpu
2. From what I understand  the real problem everybody has is the lack of a
portable and stable programming environment that does not require to
re-write large parts of complex codes into exotic languages
3. This is one of the stated goals of the MaX EU Centre of Excellence. How
well we are doing ... we'll see in twp years or so.
QE-gpu and Sirius use rather different approaches: the former strives to
minimize the differences between the GPU-enabled and the normal versions of
QE.

Disclaimer: I know very little about GPUs

Paolo


On Thu, Aug 22, 2019 at 10:13 AM Jonathan Moussa <jemoussa at vt.edu> wrote:

> Hi, I'm a software scientist at the Molecular Sciences Software Institute [
> https://molssi.org], which is an NSF-funded center tasked with providing
> software-related support to the computational chemistry community. Right
> now, I am contacting developers of popular open-source domain software to
> identify software needs related to increasing heterogeneity in HPC (e.g.
> mixed precision, GPUs, and increased shared-memory parallelism) by
> gathering their opinions. In your case, I'm inquiring about Quantum
> ESPRESSO. I'm emailing the developer list rather than a specific person
> because that's what your website recommends for contacting developers.
>
> My generic set of questions to initiate discussion are:
> 1. Is Quantum ESPRESSO ready for use on more heterogeneous HPC machines?
> 2. Is Quantum ESPRESSO having any problems preparing for heterogeneous HPC
> that could be solved by incorporating external software (available or
> hypothetical)?
> 3. Is Quantum ESPRESSO doing anything especially well that other
> developers would benefit from if it were extracted and packaged as more
> modular/portable software?
>
> Are new performance-related developments such as the GPU version
> coordinated in some way with the QE-Sirius project to improve the
> modularity of the code? That's the impression I get from browsing some of
> the slides on the GPU version, but I'm not very familiar with the
> relationship between these efforts.
>
> --
> Jonathan Moussa
> Software Scientist
> Molecular Sciences Software Institute (MolSSI) @ Virginia Tech
> _______________________________________________
> developers mailing list
> developers at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/developers
>


-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20190822/7ea1428e/attachment.html>