In a remarkable demonstration of computational prowess, a group of scientists from the Swiss National Center of Competence in Research NCCR MARVEL embarked on an extensive computational run at the recently inaugurated Alps supercomputer. This powerful machine, located at the Swiss National Supercomputing Center (CSCS), officially launched its operations on September 14, 2024. With roots embedded in the picturesque Swiss Alps, the supercomputer embodies cutting-edge technology and represents a significant stride forward in the realm of materials science.
The incredible feat achieved by the NCCR MARVEL team, which spanned over 20 hours fueled by an interminable supply of coffee, not only showcased the power of the Alps supercomputer but also highlighted the maturity of Swiss-developed computational tools. This research undertaking was vital in establishing the Alps as a premier facility capable of executing complex simulations that can lead to transformative advances in materials discovery.
On the nights of July 17 and 18, 2024, scientists from the Laboratory for Materials Simulation (LMS) at PSI, led by Giovanni Pizzi and Nicola Marzari, engaged in what they termed a “hero run.” This special allocation gave them exclusive access to the full capacity of the Alps machine for a continuous period. In computational terms, a hero run enables researchers to maximize a system’s capabilities without interruptions, allowing for an exploration of its limits.
The team’s objective was to harness AiiDA—an open-source framework designed for automating complex calculations—by coupling it with the Alps supercomputer to perform high-throughput calculations. Such calculations are crucial for evaluating thousands of potential materials, assisting researchers in narrowing down options for practical applications like battery technologies.
As Pizzi’s group prepared for the run, they aimed to demonstrate AiiDA’s potential to efficiently manage numerous simultaneous workflows, a requirement for any high-throughput computational environment. They initiated the run under the watchful eye of a dedicated scheduling system that efficiently distributed computation jobs to the supercomputer’s impressive architecture featuring 2033 NVIDIA Grace Hopper nodes, equipped with thousands of GPUs and CPU cores.
Throughout this epic marathon of computation, the brilliance of AiiDA became evident, as it facilitated job management seamlessly. By sending input files to the Alps machine and monitoring progress remotely, the team efficiently executed tasks even during periods of rest. Around 3 AM, the scientists took a short break, knowing AiiDA would handle ongoing jobs. Such coordinated efforts exemplify the potential of computational tools to enhance human abilities.
Upon completion of the run, Pizzi noted the exceptional stability of the Alps machine, evidenced by a stunning 99.96% utilization rate during their operations. This result is not only a testament to the robustness of both the hardware and software involved but also serves to underline the overarching goals of NCCR MARVEL in the quest for efficient and advanced materials discovery.
During the computation phase, the team focused on the electronic properties of around 20,000 crystal structures, all sourced from the comprehensive AiiDA database. The meticulous selection process involved choosing medium-sized structures of approximately 40 atoms to ensure effective utilization of the supercomputer’s formidable capabilities. This strategy was essential in extracting substantial insights into the materials’ potential applications and performance.
The simulations were intended to gauge various properties such as magnetism, geometric configurations, and electronic behaviors. Moreover, the researchers aimed to test novel pseudopotentials, enhancing the accuracy of their findings. These outputs hold significant scientific value; once fully processed, they will be made publicly available as FAIR data through the NCCR MARVEL’s Materials Cloud platform, enriching the global community’s access to knowledge.
The results yielded from this remarkable computational effort underscored not only the potential of the Alps supercomputer but also the innovative spirit driving computational materials research in Switzerland. By compressing in a single day what would typically take larger supercomputing initiatives a year to accomplish, the team demonstrated an unprecedented level of efficiency and capacity.
The findings from this marathon run, equivalent to 800,000 GPU hours of computation, set an inspiring benchmark for future endeavors in the field of materials science. As global challenges regarding material sustainability and energy storage become increasingly pressing, the relentless pursuit of knowledge through advanced computational methods stands to play an integral role in navigating these complexities.
Through collaborative efforts and the powerful capabilities of sophisticated technology, researchers are poised to unlock new avenues for exploration and discovery, ultimately driving innovations that will shape our understanding of materials for years to come.