Running GROMACS Efficiently on LUMI Workshop 2024

The “Running GROMACS Efficiently on LUMI Workshop 2024” ( bioexcel.eu/events/running-gromacs-efficiently-on-lumi-workshop-2024 ) was held online on the 24th and 25th of January 2024. The workshop came into being after brainstorming between GROMACS developers at PDC and the KTH Royal Institute of Technology (KTH) and the application expert teams at the CSC IT – Center for Science (CSC) in Finland about how to address the challenges of onboarding new users to the LUMI GPU partition (which is usually referred to as LUMI-G).

The workshop was run in collaboration with the BioExcel Centre of Excellence for Computational Biomolecular Research (BioExcel CoE) and gave practical tips on how to run GROMACS simulations efficiently on LUMI-G, that is, on AMD GPUs. The participants also learned how to assess and tune the performance of GROMACS. In addition, the course provided an overview of the LUMI architecture and heterogeneous parallelisation in GROMACS with special attention paid to AMD GPUs. In total, the workshop brought together 40 attendees from different European countries, with participants coming both from academia and the business/industry world.

There were two days full of lectures and hands-on sessions with a total of eight lecturers/trainers from CSC and KTH. The workshop started with an introduction to the LUMI architecture by Rasmus Kronberg from CSC ( zenodo.org/records/10683366/files/lumi-arch.pdf ) followed by a brief introduction to GROMACS by Alessandra Villa from PDC ( zenodo.org/records/10683366/files/Short_Intro_GROMACS_LUMI.pdf ). Then, Szilárd Páll from PDC addressed diverse aspects of GROMACS parallelisation/heterogeneous/GPU algorithms in depth ( zenodo.org/records/10683366/files/LUMI-WS_Szilard-Pall_GROMACS-parallelization.pdf ) and Andrey Alekseenko of the Department of Applied Physics at KTH spoke about AMD GPU support in GROMACS ( zenodo.org/records/10683366/files/AMD-GPU-support-in-GROMACS.pdf ). After the lunch break, Rasmus Kronberg from CSC described how to run programs on LUMI ( zenodo.org/records/10683366/files/lumi-batch.pdf ) and the first hands-on session started.

The following day opened with a lecture by Szilárd Páll on assessing and tuning the performance of GROMACS simulations ( zenodo.org/records/10683366/files/LUMI-WS_Szilard-Pall_Asessing_and_tuning_GROMACS_performance_on_heterogeneous_systems.pdf ), which was then followed by three consecutive hands-on sessions about GPU-accelerated simulations, scaling GROMACS across multiple GPUs, and ensemble parallelisation across multiple GPUs ( zenodo.org/records/10683366/files/Running_molecular_dynamics_simulations_with_GROMACS_on_LUMI-HackMD.pdf ).

Due to the similarity between the Dardel and LUMI architectures, the materials from this workshop will be very useful to Dardel users. If you would like to learn more, you can find the lecture material at zenodo.org/records/10683366  and the batch scripts and reference log files for the hands-on exercises can be downloaded from github.com/Lumi-supercomputer/gromacs-on-lumi-workshop/releases .