How would the two top German supercomputers: the Jülich supercomputer, JUWELS, and the Karlsruhe high-performance computer, HoreKa, perform in the computationally intensive training of Artificial Intelligence (AI) algorithms? A research team of Helmholtz AI consultants from the Jülich Supercomputing Centre (JSC) at Forschungszentrum Jülich and the Steinbuch Centre for Computing (SCC) at Karlsruhe Institute of Technology (KIT) set out to find just that. And, they jointly submitted their results for the MLPerf™ HPC benchmarking suite, which determines the performance of classical supercomputers when training neural networks. 

"With the MLPerf HPC Benchmark, we were able to use over 3000 GPUs simultaneously to accelerate machine learning. In total, more than 100 quadrillion computing operations per second were executed," explains Stefan Kesselheim, head of the Helmholtz AI consultant team at JSC. It took just two minutes to train a large climate model. On the predecessor of the JUWELS supercomputer, the task would have taken several hours.

"Not only did we use these benchmarks to better understand current systems in order to improve future systems but we also tested tools to inform users of the carbon footprint of each individual computing job," says Daniel Coquelin, a Helmholtz AI consultant at KIT. While striving for performance, it is vital to balance the environmental costs of such large-scale measurements. With JUWELS and HoreKa ranking among the top 15 on the worldwide Green500 list of energy-efficient supercomputers, the high performance computing resources in Helmholtz AI are both computationally and energy efficient. 

More information:

Press release from Forschungszentrum Jülich: Measuring the AI Performance of Europe's Fastest Supercomputer