Top European Supercomputer Shines Brighter with 70-Petaflops Booster Module


Reading time ( words)

The Jülich Research Center, long counted among high-performance computing royalty in Europe, is adding a new gem to its crown.

The center, known in German as Forschungszentrum Jülich, is extending its JUWELS supercomputer system with a booster module. Designed to provide the highest application performance for massively scalable workloads, the system is scheduled to be deployed next year.

Developed in cooperation with Atos, Mellanox, ParTec and NVIDIA, the booster module is powered by several thousand GPUs and is expected to provide a computational peak performance of more than 70 petaflops once fully integrated, up from its current level of 12 petaflops.

Building HPC Systems Block by Block

The original JUWELS supercomputer was built following a modular supercomputing architecture. The first module of the system, which started operation last year, was designed from the start to have multiple complementary modules added to it.

This innovative way of building HPC and high-performance data analytics systems follows a building-block principle. Each module is built to meet the needs of a specific group of applications. The specialized modules can then be dynamically combined as required, using a uniform system software layer.

“The modular supercomputing architecture makes it possible to integrate the best available technologies flexibly and without compromise,” said Thomas Lippert, director of the Jülich Supercomputing Center.

NVIDIA_supercomputer.jpg

Powered by the latest-generation NVIDIA GPUs with 200 Gb/s HDR InfiniBand from Mellanox, the new booster module is by far the largest in the JUWELS cluster. It brings with it a throughput and floating-point performance-optimized architecture designed for large-scale simulation and machine learning workloads.

More Brain Power

One of the initiatives the booster will fuel is the Human Brain Project. This flagship project is led by brain scientist Katrin Amunts at Jülich’s Institute of Neuroscience and Medicine and connects the work of some 500 scientists at more than 130 universities, teaching hospitals and research centers across Europe.

Created in 2013 by the European Commission, the project works to build a unique European technology platform for neuroscience, medicine and advanced information technologies — and supercomputing is integral to this.

The project’s scientists gather, organize and disseminate data describing the brain and its diseases at an unprecedented scale. To integrate all this data, the Jülich team and international collaborators are building the most detailed brain atlas to date. That’s no easy task — the human brain, with about 86 billion neurons and 100 trillion connections, is one of the most complex systems known to man.

They do this by analyzing images of thousands of ultrathin histological brain slices. The JUWELS cluster helps to solve many of the memory and performance bottlenecks on the way to reconstructing these slices into a 3D computer model.

Cracking Climate Change

The new JUWELS booster is also enabling insights into the processes behind climate change, which poses major risks for the Earth’s ecosystems.

Jülich’s “Simulation Laboratory Climate Science” provides support for an internal community of scientists who are already using JUWELS for the numerical modeling of the Earth’s systems. The booster will aid its efficiency as researchers delve into this grand challenge for the 21st century.

 

Share

Print


Suggested Items

'Eyes' for the Autopilot

07/05/2019 | TUM
Automatic landings have long been standard procedure for commercial aircraft. While major airports have the infrastructure necessary to ensure the safe navigation of the aircraft, this is usually not the case at smaller airports.

DARPA Tests Advanced Chemical Sensors

05/03/2019 | DARPA
DARPA’s SIGMA program, which began in 2014, has demonstrated a city-scale capability for detecting radiological and nuclear threats that is now being operationally deployed.

Researchers Selected to Develop Novel Approaches to Lifelong Machine Learning

05/07/2018 | DARPA
Machine learning (ML) and artificial intelligence (AI) systems have significantly advanced in recent years. However, they are currently limited to executing only those tasks they are specifically designed to perform and are unable to adapt when encountering situations outside their programming or training.



Copyright © 2019 I-Connect007. All rights reserved.