The Challenge

To prepare for future exascale computing systems, it is essential to adapt key scientific codes to the evolving HPC landscape. The Quantum ESPRESSO suite, widely used for quantum materials simulations, must fully use emerging architectures like EUPEX, featuring High Bandwidth Memory, ARM SVE instructions, and multi-core nodes with up to 160 cores. Meeting this challenge requires deep technical collaboration between application developers and HPC experts, aiming to maximise data throughput, parallelisation efficiency, and hardware-specific optimisation.

The Background

Quantum ESPRESSO is a leading open-source suite for simulating materials at the quantum level. It has introduced groundbreaking methods such as Car-Parrinello molecular dynamics and Density-Functional Perturbation Theory. With its community-driven development, Quantum ESPRESSO serves a large user base across fields like chemistry, physics, and engineering. As European HPC systems become more complex and powerful, adapting such codes to new memory hierarchies, vector capabilities, and core structures is essential for maintaining scientific competitiveness.

The Solution

The solution began with the extraction of a mini-app from Quantum ESPRESSO’s core FFT routines. This reduced complexity and focused optimisation efforts. The teams then tested performance on new architectures, especially those featuring HBM and ARM SVE. Two critical code optimisations were developed to improve vectorisation and memory usage. These changes were evaluated on emerging processors, enabling collaboration between software and hardware teams to fine-tune code paths and assess performance trade-offs in a controlled environment.

The Achievement

This collaborative effort delivered tangible improvements. Experiments on Intel Sapphire Rapids with HBM showed a 33% performance boost compared to DDR5. The new code optimisations reduced execution time by up to 21%, depending on architecture and compiler. The mini-app enabled SVE vectorisation, confirming readiness for ARM-based processors. These outcomes demonstrate the effectiveness of targeted co-design in making applications future-ready and extracting maximum performance from specialised hardware.

The Impact

Quantum ESPRESSO is essential for predicting materials properties across industries such as electronics, catalysis, and metallurgy. As one of the most used codes in its field, its performance is critical to researchers and engineers. Efficient adaptations of this code help guide HPC system procurement, shape future architectures, and reduce time-to-discovery. The mini-app approach also sets a precedent for other scientific domains, supporting more cost-effective and sustainable use of HPC infrastructure across Europe.

Key Takeaways

The mini-app format simplified collaboration and testing;

• Up to 33% gain from using HBM over DDR5;
• Up to 21% gain from code-level optimisations;
• Demonstrated effective co-design methodology;
• Ensured Quantum ESPRESSO’s future compatibility with exascale systems.