High Performance Computing System

High-performance computing (HPC) systems are crucial for tackling computationally intensive scientific problems and increasingly, for powering large-scale AI applications. Current research focuses on optimizing HPC system design for AI workloads, including efficient resource allocation, novel cooling strategies (like liquid cooling), and improved performance modeling for various AI architectures (e.g., transformers, large language models). This involves developing advanced techniques like machine learning-driven auto-tuning, optimized communication frameworks, and efficient I/O management to address the unique demands of AI and scientific simulations, ultimately enabling faster, more energy-efficient, and reliable computations for a wide range of scientific disciplines.

Papers

May 13, 2022

Deep Reinforcement Learning for Computational Fluid Dynamics on HPC Systems
Marius Kurz, Philipp Offenhäuser, Dominic Viola, Oleksandr Shcherbakov, Michael Resch, Andrea Beck
Reinforcement Learning Deep Reinforcement Learning Computational Fluid Dynamic High Performance Computing System Eddy Viscosity

April 28, 2022

Predicting batch queue job wait times for informed scheduling of urgent HPC workloads
Nick Brown, Gordon Gibb, Evgenij Belikov, Rupert Nash
High Performance Computing System Heavy Workload Customer Side Queue

March 2, 2022

Hyperparameter optimization of data-driven AI models on HPC systems
Eric Wulff, Maria Girone, Joosep Pata
Data Driven Hyperparameter Optimization High Performance Computing System Exascale Computing Large Scale Hyperparameter Optimization

High Performance Computing System

Papers

Deep Reinforcement Learning for Computational Fluid Dynamics on HPC Systems

Predicting batch queue job wait times for informed scheduling of urgent HPC workloads

Hyperparameter optimization of data-driven AI models on HPC systems