Milestone in Europe in the development of a new generation of flexible, modular and energy efficient supercomputers

Christoph Hammerschmidt writes on the EE News Europe website about a new innovative supercomputer that is helping Europe catch up with other global developments.

 

European supercomputer shines with energy efficiency

Europe is catching up in developing innovative supercomputer architectures. One example of this is the new high-performance computer that is currently being launched at the Jülich Supercomputing Centre (JSC).

JUWELS (Jülich Wizard for European Leadership Science) is regarded as a milestone in the development of a new generation of flexible, modular supercomputers that target an extended range of tasks, from big data applications to computational simulations. With his first module it qualified as number 1 of the German computers for the TOP500 list of the fastest computers in the world.

Supercomputers help to understand complex interrelationships in research, for example in climate research or neuroscience. But they are also increasingly being used in many other areas. In addition to classical applications such as simulations in engineering, physics or chemistry, supercomputers are increasingly being used for other tasks involving the evaluation of large amounts of data or machine learning and artificial intelligence (AI). JUWELS has a modular design to meet such different requirements. Thomas Lippert, Director of the Jülich Supercomputing Centre, sees this approach as the key to an affordable and energy-efficient technology with which the coming exascale systems in particular can be realised.

Lippert’s concept of an adaptable design – also known as “Smart Exascale” – has developed into a comprehensive European project in recent years and has been turned into reality in the EU research project DEEP. Since 2011, experts from 16 European partners have been working on DEEP projects funded by the EU.

The modular concept provides for a supercomputer consisting of several specialized modules that can be dynamically and flexibly combined as required via software. The cluster module delivered in spring 2018 by French IT company Atos together with the software specialists of the German company ParTec is equipped with Intel Xeon 24-core Skylake CPUs. This results in a theoretical peak performance of up to 12 petaflops, roughly equivalent to the computing power of 60,000 PCs. The nodes are interconnected by a Mellanox InfiniBand high-speed network with transfer rates of up to 200 Gb/s.

Another special feature of the module is its particularly energy-efficient hot water cooling. This form of cooling makes it possible to cool the majority of the waste heat with hot water directly with the outside air without additional cooling generators.

Next year a booster module for massively parallel applications is to be added, which will multiply the computing power of JUWELS. The booster is equipped with a large number of very energy-efficient computing cores, which are connected by a particularly powerful and fast network. The booster works similar to a turbocharger: complex parts of the code that cannot be efficiently calculated on a large number of processors are executed on this cluster. Program parts that can be processed in parallel with maximum efficiency can be dynamically outsourced to the booster module.

Even the cluster module achieved a computing speed of 6.2 petaflops in initial test runs with the Linpack benchmark, although it is designed for flexibility and universality rather than speed. This makes the system the fastest German computer in 24th place on the TOP500 list of the fastest computers in the world, which was recently presented at the International Supercomputer Conference ISC.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

w

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.