Experts explain that exascale marks a new level of supercomputing speed. Our personal computers, equipped with fast graphics cards, can perform up to a trillion (10¹²) operations per second, while exascale machines can achieve a quintillion (10¹⁸) or more.

To put that into perspective: if every person on Earth could perform one operation per second, it would take nearly four years for all of humanity to do what a supercomputer can do in just one second. This immense speed enables advances in many fields—from precision medicine and climate research to materials science—but only if the software knows how to harness that power.

The team behind LimitX, along with the Ruđer researchers, includes the Jülich Supercomputing Centre (JSC) in Germany and the French Alternative Energies and Atomic Energy Commission (CEA). They spent a year developing this software within the Inno4scale European project, valued at €4.5 million.

The project is led by the Barcelona Supercomputing Center, with partners including Swiss SCAPOS, Germany's High-Performance Computing Center Stuttgart (HLRS), and the PRACE network.

What is it about, and how does the new software work?

Until now, quantum protein simulations used numerical methods based on quantum mechanics to calculate electron distributions and system energy, typically via computationally intensive iterative processes. This approach is often slow and inefficient due to the need to calculate a large number of different parameters, but LimitX changes that.

“The basic idea of the new method is to try to ‘guess’ a good initial solution,” explains Dr. Davor Davidović, Head of CIR.

“For example, imagine someone leaves you on a road and tells you to find your way to Rome as quickly as possible. Without any extra information, it would take you a long time, relying on trial and error. But if you know you’re in the north and need to go south, the number of attempts drops significantly, and you'll reach your destination much faster.”

Similarly, before starting the main simulation, LimitX learns from previous examples and then predicts the most likely electron distribution using machine learning and neural networks. Thanks to this initial estimate, the computer requires far fewer steps to reach the same accurate result, cutting down computation time from several days to just a few hours, or even minutes.

This kind of acceleration unlocks many possibilities. Pharmaceutical companies could test hundreds of thousands of molecules in a week instead of just a few dozen. Researchers and engineers working on new materials could discover new compounds much faster. As exascale computers reach unprecedented speeds, smart software like this ensures that power is fully utilized.

Dr. Davor Davidović emphasizes: “As exascale computers arrive, the real question becomes how to use them wisely. Without new and advanced algorithms and software, even the fastest hardware can’t find the right path on its own.”

Next step: Pilot testing on EuroHPC

In the coming months, the CIR team is preparing LimitX for pilot testing on real exascale computers available to academia and industry through the EuroHPC supercomputing network.

“If current preliminary results are confirmed, the proposed computational method could soon become part of standard toolkits for quantum simulations used by universities, research institutes, and pharmaceutical companies across Europe,” explains Dr. Jurica Novak from CIR.

The LimitX team from IRB includes Dr. Jurica Novak, Dr. Abhiram Kaushik Badrinarayanan, and Dr. Davor Davidović, with other team members from the Jülich Supercomputing Centre, Dr. Edoardo di Napoli, Dr. Xinzhe Wu, and from France’s CEA Dr. Luigi Genovese.