Seminar on Efficient Programming of HPC Systems - Frameworks and Algorithms

Wednesday 8 Jul 2026, 09:00 → 18:00 Europe/Berlin

Hörsaal (Ground Floor) (MPCDF)

09:00 → 09:05 Introduction

Speaker: Erwin Laure (MPCDF)

09:05 → 09:30 HPX - A modern C++ task parallelization framework

Speaker: Guillermo Marcos Lara

09:30 → 09:55 Portable GPU Programming with OpenMP Target Offloading

Speaker: Denys Myshak

09:55 → 10:20 Julia for High-Performance Computing

Speaker: Daniel Singh

10:20 → 10:35 Break

10:35 → 11:00 Seminar Paper on HPC Storage and Lustre

Speaker: Daymon Schodits

11:00 → 11:25 IO-Aware Attention Across GPU Generations: How FlashAttention Tracks the Moving Bottleneck

Speaker: Tom Osterfeld

11:25 → 11:50 Balancing Performance and Accuracy: Mixed-Precision Algorithms for Linear Algebra in HPC

Speaker: Felix Weißleder

11:50 → 12:45 Lunch break

12:45 → 13:10 Alpaka

Speaker: Bora Uygar Özyurt

13:10 → 13:35 An MLIR-Based Approach to HPC Portability

Speaker: Yannick Schürmann

13:35 → 14:00 JAX/XLA for Accelerator Performance Portability: Compiler Mechanisms and Quantitative Evidence

Speaker: Pau Marín Roig

14:00 → 14:15 Break

14:15 → 14:40 Mitigating Load Imbalance in Molecular Dynamics through Adaptive Task Parallelism

Speaker: Audrey Kyrene Chen Kartamihardjo

14:40 → 15:05 ADIOS in High-Performance Computing Architecture and Performance Improvements for Scientific Simulations

Speaker: Thomas Krachten

15:05 → 15:30 Data Layouts on Heterogeneous Systems

Speaker: Lukas Englhauser

15:30 → 15:45 Break

15:45 → 16:10 Scaling Massive Models Efficiently: Fully Sharded Data Parallelism in PyTorch

Speaker: Sai Krishna Sriyash Kommalapati

16:10 → 16:35 Non-IEEE and Reduced-Precision Number Formats for AI and HPC

Speaker: Paul Fleischmann

16:35 → 17:00 Model parallelization strategies for inference and training

Speaker: Nils Marvin Quiring

17:00 → 17:20 Conclusions