Computational marathon matches the efficiency of existing platform with the power of new supercomputer

It took about 20 hours and a number of espresso for a group of scientists from the Swiss Nationwide Middle of Competence in Research NCCR MARVEL to finish a computational marathon that showcased each the facility of Switzerland’s fundamental supercomputing facility, and the extent of maturity achieved by Swiss-made software program instruments for computational supplies science.

The Alps supercomputer, which simply turned operational with its official inauguration on September 14, 2024, is without doubt one of the world’s strongest supercomputers. It’s managed by the Swiss Nationwide Supercomputing Middle (CSCS) and it consists of a geo-distributed infrastructure primarily situated within the Lugano knowledge middle.

Through the acceptance section, CSCS allowed entry to Alps to chose analysis teams, and among the many first with this chance have been members of the NCCR MARVEL, particularly Giovanni Pizzi’s group, a part of the Laboratory for Supplies Simulation (LMS) at PSI headed by Nicola Marzari, that makes use of computational methods to search for new supplies for a lot of functions.

Over the course of someday and one evening on July 17 and 18, a group together with Marnik Bercx, Michail Minotakis and Timo Reents, all from Pizzi’s group, launched into what computational specialists name a “hero run”—a time slot when a supercomputing machine is totally reserved for a single person, to make use of the complete energy of the whole machine to advance their very own analysis, and show their functionality of effectively exploiting the immense computational energy of the complete system.

The PSI group wished to match the facility of the Alps supercomputer with AiiDA, an open-source software that helps supplies scientists automate the lengthy and complicated calculations required to simulate the properties of supplies—both current ones or these nonetheless ready to be found.

Specifically, they interfaced AiiDA and Alps to run high-throughput calculations, the place hundreds of various supplies buildings saved in a database are calculated in parallel. It’s the type of computational experiment that enables, for instance, the choice of potential new battery supplies out of hundreds of identified chemical compounds, serving to experimentalists to focus their efforts on probably the most promising ones.

“We wanted to show that AiiDA can fill up all the nodes of a supercomputer with near-exascale performance for many hours and fully exploit the power of the machine while handling, running and maintaining many separate workflows simultaneously, which is necessary for high-throughput computations,” explains Bercx.

The run was managed remotely, with the AiiDA software program put in on a PSI server, and used to organize all enter information of the calculations to be carried out. The precise computations have been executed utilizing an enhanced model of the broadly used Quantum ESPRESSO computed code for supplies simulations, powered by the Sirius library—developed inside NCCR MARVEL at CSCS—that enables for the optimum exploitation of the good computing energy supplied by graphical processing models (GPUs) of Alps, and implements novel algorithms to considerably enhance the simulation success fee.

When the scientists bought the inexperienced gentle from the CSCS workers round midday on the chosen date, they began sending enter information to the Alps machine, the place they have been submitted to a scheduling software program that distributed the roles among the many 2033 NVIDIA Grace Hopper nodes (together with 8,132 GPUs and 585,504 CPU cores) that have been granted for the hero run and queued them. On the opposite aspect of the connection, AiiDA was monitoring every job so that when it was completed, the information may very well be retrieved, parsed, and saved in AiiDA, and new calculations may very well be then submitted.

In a short time after beginning the run, AiiDA may fill the entire Alps supercomputer with jobs, totally exploiting its excellent computational capabilities. Round 3 AM, the group understandably wanted a brief nap, and relied on AiiDA to proceed getting ready and submitting new jobs of their absence. The run efficiently ended round 9 AM on the second day.

“All went smoothly, and the number of available nodes was remarkably stable during the entire hero run, which speaks to the quality of the infrastructure” says Bercx. The 99.96% utilization of a near-exascale machine is totally exceptional and fairly unprecedented—very a lot reaching the objectives of the MARVEL NCCR devoted to computational supplies discovery enabled by such capabilities and infrastructure.

In the long run, the group managed to finish nearly 100,000 calculations, similar to single runs of Quantum ESPRESSO, in nearly 16 hours. Extra particularly, the calculations have been concerning the properties of round 20,000 crystal buildings taken from the AiiDA database.

“We chose medium-sized structures, because Alps is so powerful that small structures would not use the computational power efficiently,” explains Minotakis. “We started with structures made out of 40 atoms, and then in subsequent submissions added slightly smaller and slightly larger structures.”

The computations have been meant to calculate the digital properties of the supplies of their floor state, discover whether or not they have been magnetic or not, and calculate their ground-state geometric configuration.

“We also had new pseudopotentials that we wanted to test, so we updated the calculations for a large fraction of the structures in the database and checked the differences with previous calculations” says Reents. All the outcomes will quickly be revealed as FAIR and open knowledge, and uploaded to the Supplies Cloud, the net knowledge sharing platform of NCCR MARVEL, to broaden the MC3D database of inorganic 3D crystal buildings.

Along with the good scientific worth of those simulations, the run demonstrated the effectivity and stability of AiiDA, which may seamlessly fill the whole capability of an exascale machine.

“The performance of the new Alps machine is outstanding, even more so when combined with the high-throughput capabilities of AiiDA. It is impressive that we could compress in less than a day the equivalent computing power granted for one full year to large supercomputing projects at CSCS, equivalent to approximately 800,000 GPU hours of computation on the previous-generation CSCS supercomputer Daint,” says Pizzi.

Supplied by
Nationwide Centre of Competence in Research (NCCR) MARVEL