Testing on LUMI-C delivers superior parallel performance

LUMI, Europe’s leading supercomputing system, has garnered attention for its exceptional computational power and efficiency. Jiří Jaroš from Brno University of Technology in the Czech Republic has been exploring the capabilities of LUMI. Known for his contributions to biomedical ultrasound simulations and the development of high-performance computing tools, Jaroš has already discovered LUMI-C’s superior parallel performance. Jaroš’s team ran simulations on the Karolina supercomputer, another EuroHPC JU system, and applied for computing time on LUMI via the IT4Innovations Open Access Grant Competitions. As they continue to develop their codes for LUMI, the team is witnessing first-hand the impressive capabilities of LUMI-C.

Improving treatment planning for transcranial ultrasound simulation

The advancements facilitated by LUMI-C are particularly significant for Jiří Jaroš and his team’s research endeavors, which often require extensive computational resources. Their work focuses on improving treatment planning for transcranial ultrasound stimulation, relying heavily on precise simulations to navigate the complexities of acoustic wave propagation through biological tissues.

Central to their simulation capabilities is the k-Wave toolbox, an essential software framework developed in collaboration with Bradley Treeby and Ben Cox from University College London (UCL).

– K-Wave has become one of the most widely-used tools for modeling biomedical ultrasound and photoacoustics. It has enabled key advances such as improving the resolution of in vivo photoacoustic images, predicting ultrasound fields in the body for treatment planning and safety assessment, and quantitative ultrasound imaging using waveform tomography, noted Treeby.

Image: Bradley Treeby presenting treatment planning for ultrasound brain stimulation using HPC as a service at the HPCSE 2024 conference.

Jiří Jaroš is the lead developer of the accelerated versions of k-Wave that are optimized for high-performance computing architectures such as GPUs and distributed clusters. Meanwhile, the team at UCL focuses on advancing the underlying algorithms and numerical methods to enable new modeling capabilities.

– We conducted the majority of our testing on LUMI-C, which delivers superior parallel performance compared to the Karolina supercomputer, thanks to LUMI’s newer CPUs and fewer NUMA domains, explained Jaroš. The code for LUMI-G is currently under development, he added.

Image: Jiří Jaroš and the Karolina supercomputer at IT4Innovations National Supercomputing Center.

Extremely large simulations on LUMI

Looking ahead, Jaroš envisions expanded applications of LUMI in their research, foreseeing significant advancements.

– In our case, this is the first machine equipped with AMD GPUs that we have used. Consequently, the primary benefit lies in porting our codes to this new architecture. Looking ahead, the main advantage could be the ability to run extremely large simulations that are not feasible on Karolina due to size constraints, Jaroš remarked.

Jaroš also expressed his satisfaction with the LUMI Support Team and training events:

– We are very satisfied with the support provided by LUMI. I have attended several courses on LUMI-C and LUMI-G, where I acquired all the necessary knowledge, he adds.

Author: Barbora Polakova, IT4Innovations National Supercomputing Center