In a remarkable showcase of technological prowess and scientific collaboration, a dedicated team of researchers from the Swiss National Center of Competence in Research (NCCR MARVEL) recently pushed the limits of computational science with the Alps supercomputer. Launched on September 14, 2024, this cutting-edge machine stands as a testament to Switzerland’s commitment to advancing computational materials science. Situated primarily in the Lugano data center and overseen by the Swiss National Supercomputing Centre (CSCS), Alps marks a significant milestone in global supercomputing rankings.
The supercomputer’s remarkable capabilities were put to the test during what is informally known as a “hero run,” a designated time frame in which a singular research group monopolizes the supercomputer’s full computational resources. On July 17 and 18, a dedicated team from NCCR MARVEL, led by Giovanni Pizzi, embarked on this challenge and showcased the sophisticated software developed within their research framework, further solidifying the role of Swiss innovations in tackling complex scientific problems.
The team consisted of talented researchers including Marnik Bercx, Michail Minotakis, and Timo Reents, who aimed to explore the capabilities of their open-source software AiiDA, designed to facilitate the automation of laborious calculations in materials science research. This approach contrasts with traditional methods, which may be prohibitively time-consuming and prone to human error. Their goal was not only to utilize the Alps supercomputer’s capabilities fully but also to demonstrate AiiDA’s potential in managing multiple workflows concurrently—something crucial for high-throughput calculations involving extensive material structures.
Through a meticulous interfacing process, they were able to synchronize the AiiDA software with the Alps supercomputer, launching high-throughput calculations that would permit the examination of thousands of material structures simultaneously. By tapping into extensive databases of chemical compounds, the team targeted the discovery of new materials, particularly in the field of batteries—a crucial endeavor given modern energy demands and the need for sustainable technology solutions.
The intersection of AiiDA and the Alps supercomputer highlighted a new paradigm in computational materials science. The team utilized an enhanced version of Quantum ESPRESSO, a well-known computational code in the field, equipped with a specific library called Sirius. This combination ensured optimization for graphical processing units (GPUs), amplifying the supercomputer’s processing capabilities and substantially improving the success rate of simulations.
Commencing at noon on their chosen day, the team initiated their computational sequence to fill the 2033 NVIDIA Grace Hopper nodes. Each node, equipped with a staggering number of GPU and CPU cores, became instrumental in executing the planned calculations. Significantly, AiiDA assumed a central role, managing submissions and smoothly transitioning between tasks, even autonomously continuing operations when the human team required rest—a feature indicative of its robust design.
Approximately 16 hours later, the team concluded their ambitious computational run, achieving near-exascale performance with an unprecedented utilization rate of 99.96%. This efficiency not only surpassed expectations but exemplified the capabilities of both AiiDA and the Alps supercomputer in performing high-throughput computations seamlessly. The successful management of nearly 100,000 calculations—focused predominantly on examining the properties of roughly 20,000 distinct crystal structures—stands to advance knowledge in materials science significantly.
The calculations aimed to determine various properties, including electronic states and magnetic characteristics of the materials examined. Alongside these explorations, the team tested new pseudopotentials, updating critical datasets and assessing variances with prior calculations. As a culmination of their efforts, the findings will be registered as FAIR (Findable, Accessible, Interoperable, Reusable), and shared broadly through the Materials Cloud, enriching the communal database of inorganic 3D structures.
The remarkable outcomes of this endeavor reflect not just a rapid computational turnaround but also a burgeoning future for experts in materials science. As Switzerland continues to invest in supercomputing technology, endeavors like the hero run pave the way for significant advancements that can catalyze innovative applications across various industries. With the exceptional performance of the Alps supercomputer combined with the technological flexibility of AiiDA, new frontiers in material exploration beckon, reaffirming the importance of collaboration and ingenuity in scientific research. This successful initiative not only highlights Switzerland’s prowess in the field of computational research but also underscores the potential of global cooperative research efforts.