Fraunhofer ITWM at Supercomputing 2024

Pioneering Technologies for High Performance Computing and Artificial Intelligence

Press Release / November 15, 2024

The Fraunhofer Institute for Industrial Mathematics ITWM will be presenting innovative solutions for parallel programming and Artificial Intelligence at Supercomputing 2024 in Atlanta, USA, from November 18 to 22. Two Fraunhofer spin-offs will participate at the joint booth 3115 in Hall B: ThinkparQ will present its roadmap for the parallel file system BeeGFS. The spin-off UNEEC Systems will be demonstrating the STX processor and its system approach to energy efficient and cost effective compute infrastructure. The STX processor accelerates HPC simulations while keeping programming simple. Researchers from the division »High Performance Computing« will share new developments in parallel programming and other AI solutions for High Performance Computing.

NASE – Neural Architecture Search Engine

NASE is a powerful AI tool that optimizes neural network models with respect to the underlying hardware, e.g. AI accelerators, inch AISCs and FPGAs. The tool automates the search for efficient AI models that meet specific performance requirements such as speed, latency and power consumption. By considering unique hardware characteristics in the model design, NASE ensures customized neural architectures. Advanced algorithms and extensive computing resources enable the provision of ready-to-use, efficient neural networks for real-world applications.

Carme – Open Source Software Stack for AI Infrastructure Management

The open source software stack CARME bundles Fraunhofer ITWM's expertise in Artificial Intelligence and High Performance Computing and can be seamlessly integrated into existing HPC systems. Users can focus on their AI development goals without the need for experience or knowledge of workload management tools or hardware. CARME features web-based access, two-factor authentication (2FA) and integrated container technology, making it a user-friendly and robust tool for managing interactive and batch AI workloads on HPC systems.

GaspiLS – Scalable Iterative Linear Solver for Cfd and Fem Simulations

The GaspiLS scalable iterative solver library has demonstrated exceptional scalability in codes from independent software vendors and industry. With the widespread use of workstations with over one hundred cores and small HPC installations with thousands of cores, the strong scalability of this solver library is essential for engineering solutions.

GaspiLS utilizes highly scalable Algebraic Multigrid (AMG) and hybrid Incomplete-LU preconditioners to design the optimal solver for each code. In addition, GaspiLS is open source, which encourages a collaborative approach to improving its capabilities. GaspiLS can be seamlessly integrated into any code that uses Message Passing Interface (MPI) or Global Address Space Programming Interface (GPI).

MCSS – The Next Revolution in Data Storage

For the first time, researchers from Fraunhofer ITWM will present their novel Memory Centric Storage System. The aim of the new I/O system is to eliminate the separation of memory and the various storage devices and bring them together in a distributed I/O system under a standardized interface. This makes it possible to work with bytes and files across all devices in a flexible and highly optimized way, facilitating workflows and supporting campaign storage.

The Fraunhofer ITWM HPC team is looking forward to your visit and intensive discussions.

Press Release for download as PDF [ PDF 0,13 MB ]

Further Information on Our Topics

Multi-Target Neural Architecture Optimization

AI-Services: NASE – Neural Architecture Search Engine

We support you in designing and integrating your optimal, individual neural network.

GaspiLS

GaspiLS is a scalable and industry proven linear solver library for the exascale age and trimmed to achieve optimal scalability.

Carme

With the open source multi-user software stack Carme, several users can manage the available resources of a computing cluster.

Energy-Efficient Supercomputer Systems

The Stencil and Tensor Accelerator (STX) relies on hardware-software co-design and open source to strengthen key technologies such as system design in Europe.

BeeGFS

BeeGFS (also known as Fraunhofer Parallel Filesystem) is a parallel cluster file system, developed with a strong focus on performance and designed for very easy installation and management.

GPI

GPI allows truely asynchronous and parallel communication of all threads and achieves optimal overlap of communication by computation.

GaspiCxx

GaspiCxx is a programming interface for the C++ programming language developed for the GPI-2 communication library.

Quantum Computing

In this focus, we are working on new algorithms and software for the use of Quantum Computing. The interplay between Quantum Computing and High Performance Computing is interesting.

Seismic Imaging

The seismic reflection method computes an image of the Earth's subsurface, which enables geologists to identify the locations of oil and natural gas fields.

Press Release