-
Research
Computer power for cutting-edge research
Research
Computer power for cutting-edge research
Expert on supercomputers: Dorian Krause
show
The Jülich supercomputer JUWELS has shifted up several gears at once. By putting the booster module into operation, it is now Europe’s fastest computer. This enables more complex simulations.
3 questions for ...
... Dr. Dorian Krause, Head of the High-Performance Computing Systems division at the Jülich Supercomputing Centre (JSC).
Dr. Krause, how does science benefit from the new booster of the Jülich Wizard for European Leadership Science, or JUWELS for short?
The enormous power of the booster makes it possible to solve even significantly larger problems than before, that is, to carry out more demanding simulations. Simulations are a cornerstone of research, whether it be climate, energy, life sciences, environment, materials or technology. They are becoming increasingly complex and, with increasing frequency, large amounts of data are being incorporated or are arising. In addition, artificial intelligence methods are becoming ever more important. The JUWELS system is very well suited to all these diverse tasks thanks to its cluster-booster architecture, which emerged from Jülich research work.
Image above: Expert on supercomputers: Dorian Krause
1quadrillions of computing operations per second (73 petaflop/s) can be performed by the JUWELS booster, which is roughly equivalent to the performance of over 300,000 modern PCs. The 12 petaflop/s computing power of the cluster are added to that.
JUWELS is considered a model for the European exascale supercomputer, which is scheduled to go into operation in 2023 with at least twelve times the computing power of JUWELS. Why?
JUWELS is designed in the same way as we expect from the exascale computer: its modular architecture, the design of its computing nodes, the network, the infrastructure and the cooling can be transferred to an exascale computer while keeping costs and energy consumption reasonable.
In view of quantum computers or computers that work in a way modelled on the human brain, aren’t digital supercomputers an outdated model?
No, not at all. Quantum computers and neuromorphic computers are the object of research and will increasingly become research tools of their own. The modular JUWELS system is prepared to integrate quantum computers to facilitate hybrid computing.
More about the JUWELS supercomputer (in German): fzj.de/juwels
The water of the earth in the computer
In Germany, 2018 went down in history as a drought year. The impact was felt in many areas: crop failures, tree death, restrictions in the drinking water supply, and low water levels of rivers which limited the transport of diesel and other raw materials. Using a unique computer model called the Terrestrial System Modelling Platform (TSMP), researchers can predict, among other things, over which time periods water resources may recover from such droughts under uncertain precipitation coditions.
“Our simulations with the TSMP require very large computing power, especially when it comes to high resolution: for example, it has not yet been possible to generate such invaluable water resource forecasts – including an estimate of how uncertain the respective forecast is – individually for each agricultural plot across Germany. With the new computing power of the JUWELS system, this is about to change,” explains Jülich agrosphere researcher Prof. Stefan Kollet.
By using JUWELS, Stefan Kollet wants to make more accurate forecasts of the water resources of forests and other landscapes. Strictly speaking, the TSMP is a set of interconnected computer models that together simulate the entire water cycle in and over land masses, from groundwater to the upper layer of the atmosphere. One of these models, ParFlow, benefits particularly from the new JUWELS booster. ParFlow can simulate surface and ground water flows, while also taking into account the human influences such as groundwater pumping and irrigation. “With the booster module, we will for the first time be able to perform ParFlow simulations for Germany and Europe with the required fine resolution of, for example, individual hill slopes or river corridors,” says Kollet. ParFlow is designed to run on graphics processing units (GPUs). This means that the software includes many computationally intensive tasks that can be well subdivided into computing steps and then processed in parallel on the GPUs.
This is precisely the strength of the new booster module: thanks to the more than 3,700 GPUs in its 936 cleverly linked computing nodes, it can deliver its massive computing power in an energy-efficient manner.
Unlike ParFlow, the atmosphere model used in TSMP is not parallelised for the booster’s GPU-based architecture, which would require time-consuming coupling of a different version of the atmospheric model to the TSMP. Therefore, the current version of the model will continue to be computed on the JUWELS cluster with its conventional processors, making best use of the heterogeneous architecture of JUWELS: “For different requirements in particular, as with the different component models of the TSMP, the JUWELS system with its modular cluster booster architecture offers perfect conditions,” says Kollet.
Clusters and boosters together can also be used in other scientific fields – to varying degrees, depending on the requirements. “The modularity of the JUWELS system is our answer to the increasingly complex and diverse requirements of application codes on supercomputers. It fits perfectly with our mission to provide cutting-edge European computing resources for a wide range of applications and research areas,” says the Head of the JSC, Prof. Thomas Lippert.
Texts: Frank Frick
© 2022 Forschungszentrum Jülich