SciComp Members Present at SIAM PP26 in Berlin

March 5, 2026.

SciComp members participated in the SIAM Conference on Parallel Processing for Scientific Computing 2026, held in Berlin from 3–6 March 2026. The conference is one of the leading international forums for advances in high-performance computing (HPC), numerical algorithms, and large-scale scientific simulations.

SciComp researchers contributed several presentations highlighting the group’s work across machine learning–accelerated simulation, numerical relativity, and performance optimisation for modern heterogeneous hardware.

Contributed Talks

Four members of the group delivered contributed talks:

• Tobias Weinzierl delivered a talk titled “Are Hardware Trends Bad News for Task Parallelism, Hierarchical Models and AMR in Black Hole Simulations?”. The presentation discussed how evolving hardware architectures impact task-parallel simulation frameworks used in large-scale astrophysical modelling.
• Gokberk Kabacaoglu presented “VesNet: Machine Learning Accelerated Solver for Stokesian Particulate Suspensions”, describing a neural-network approach for accelerating the evaluation of integral operators in simulations of particulate Stokes flows. The work demonstrates how machine learning can be integrated with classical numerical methods to significantly accelerate complex fluid simulations.
• Anne Reinarz talked about “Multiscale Modeling of Nonlinear Behaviour in Aerospace Composites“. Her talk introduced a multiscale solver designed to efficiently simulate nonlinear behaviour in aerospace composites, which often feature complex geometries, material heterogeneities, and localized effects such as delamination.
• Pawel Radtke, a PhD student supervised by Weinzierl, presented “Compiler-Supported Reduced Precision and AoS-SoA Transformations for Heterogeneous Hardware.” His work explores how compiler techniques can automatically transform data layouts and precision levels to improve performance on modern heterogeneous computing platforms.

Poster Presentations

SciComp researchers also presented several posters showcasing ongoing work within the group:

• Pawel Radtke — Compiler-Supported Reduced Precision and AoS-SoA Transformations for Heterogeneous Hardware
• Timothy Stokes — ExaHyPE: Recent Developments in Numerical Relativity (ExaHyPE) and Kernel Optimisation (ExaHyPE-DSL)
• Thomas Flynn — SHAREing: Performance Analysis, Training and Community for Accelerated Compute

The conference provided an excellent opportunity for SciComp members to engage with the international HPC community, exchange ideas with leading researchers, and discuss future collaborations across computational science and engineering.

Photos from the conference are included below.

Pawel’s talk	Timothy’s poster
Thomas’ poster	Pawel’s poster