11 February 2022
MaX Hackathon: porting MaX flagship codes on AMD accelerated architectures
In this hackathon, selected members from the developers' group of the MaX flagship codes will test the accelerated versions of their application on AMD based platforms.
The main objective is to preliminarily evaluate the performance and the effort required to run accelerated versions of MaX flagship codes on AMD accelerated architectures. In particular, as all accelerated versions of the MaX flagship codes are based on a combination of FORTRAN and OpenACC, the developers are going to test how the software stack installed on the available architectures supports these kind of codes. The outcome of this hackathon is likely to provide an outlook of the expected behavior - at least on a single node level - of the accelerated applications on the next coming EU pre-Exascale LUMI platform.
Technical Specifications:
a) One AMD based server provided by E4:
b) One AMD based server provided by CSC:
LUMI Pre-Early Access Platform node (eap1)
Cray Linux Environment (SLE 15 sp2) | AMD ROCm 4.3
AMD EPYC 7662 64-Core Processor | 512 GB Memory
AMD Instict MI100 Accelerator | 32 GB GPU Memory
Participants (and codes):
Pietro Delugas, SISSA, QE
Oscar Baseggio, SISSA, QE
Nicola Spallanzani, CNR, Yambo
Daniel Wortmann, FZJ, Fleur
Andrea Ferretti, CNR, Yambo
Fabrizio Ferrari Ruffino, CNR-IOM, Yambo
Andrea Marini, CNR, Yambo
Davide Sangalli, CNR, Yambo
Ivan Carnimeo, SISSA, QE
José Mª Escartín, ICN2, SIESTA
Alberto Garcia, ICN2, SIESTA
Related talks from invited speakers:
4 Feb 2022, "LUMI architecture and programming environment", George S. Markomanolis - Lead HPC Scientist, CSC
10 Feb 2022, "Accelerated linear algebra and outlook on programming models for Exascale", Piotr Luszczek, Mark Gates, Stanimire Tomov - ICL and Univerity of Tennessee
17 Feb 2022, "OpenMP GPU Offloading - a quick overview", Giacomo Rossi - Intel
Dates: 27/01 - 20/02 (probably will be extended)
Venue: fully on-line. Only invited talks are on Zoom, intra-communication between participants is on a dedicated Slack channel.
Organizers: Ivan Girotto (ICTP), Daniele Varsano (CNR) and Andrea Ferretti (CNR)
NOTES: This event is restricted to a small number of participants as the very limited amount of resources available.