September 19, 2024 – by Santina Russo
In a computational marathon, researchers harnessed the power of the ‘Alps’ supercomputer for an uninterrupted 20-hour run, demonstrating its capability to handle vast and complex computational workloads. This "hero run"—a time slot dedicated to a single user utilizing the machine's full capacity—was conducted during the ‘Alps’ acceptance phase by members of the Laboratory for Materials Simulation (LMS) at the Paul Scherrer Institute (PSI) led by Nicola Marzari, also part of the National Centre of Competence in Research (NCCR) MARVEL. The goal was to test their ability to efficiently exploit the machine’s immense computational power.
Giovanni Pizzi’s team at PSI employed AiiDA, an open-source tool that helps scientists automate the workflows for simulating material properties, to put ‘Alps’ to the test. In their high-throughput calculations, thousands of material structures stored in a database were calculated in parallel. “Everything went smoothly, and the number of available nodes remained remarkably stable during the entire hero run, which speaks to the quality of the infrastructure,” said team member Marnik Bercx. The workload ran with a sustained 99.96% machine utilization—an impressive and quite unprecedented level for a near-exascale machine. In just 16 hours, the team completed nearly 100,000 calculations, revealing the properties of around 20,000 crystal structures.
“The performance of the new ‘Alps’ supercomputer is outstanding, especially when combined with AiiDA’s high-throughput capabilities,” noted group leader Giovanni Pizzi. “It is remarkable that we were able to condense, in less than a day, the equivalent computing power previously granted for a full year of large-scale supercomputing projects at CSCS.” This successful hero run showcased the versatility and capacity of ‘Alps’, which is designed to tackle complex, data-heavy workloads across fields such as materials science, particle physics, AI research, and more.
Read the full article by MARVEL here.