Technology

AI supercomputer for energy research: Energy Efficiency Optimization department relies on NVIDIA DGX™ A100

The Department of Energy Use Optimization (EEO) is relying on an Nvidia DGX A100 supercomputer based on the Ampere architecture. The acquisition marks a milestone for research at the EEO department, particularly in the field of energy system optimization, and offers new opportunities for the development and application of artificial intelligence (AI).

The system is equipped with eight Tesla A100 Tensor Core graphics processors, six NVLink™ switches, two AMD Epyc processors, 1 TB DDR4 RAM, 15 TB NVMe SSD storage and nine Mellanox ConnectX-6 adapters. The GPUs are currently connected to each other via six NVLink™ 3.0 switches with transfer rates of 600 GByte/s (bi-directional) between all GPUs.

Rethinking optimization and efficiency in energy systems

Prof. Dr. Peter Bretschneider, Head of the Department of Energy Use Optimization, emphasizes that conventional computer systems often reach their limits when processing large amounts of data. "Especially in the field of energy system research, large amounts of data and complex algorithms are required to develop sustainable and efficient solutions. With the NVIDIA DGX™ A100, we can speed up these processes enormously and set new standards."

Wide range of applications for AI and ML

Artificial intelligence (AI) and machine learning (ML) are key technologies in the department's research activities. The DGX A100 makes it possible to train AI models with large data sets and achieve significantly higher efficiency and accuracy. Areas of application include

• Optimization of power gridsIntelligent methods and algorithms for forecasting energy demand curves, fluctuating feed-ins and optimizing energy procurement.
• Simulation of sustainable energy useDevelopment of strategies for the optimal use of electricity, gas and heating/cooling through the integration of renewable energies, as well as the integration of storage and flexibility potentials.
• Improvement of industrial processesDevelopment of strategies for the data-driven adaptation of energy system models to real processes. As well as for the calculation of optimal energy schedules taking uncertainties into account.
• Development of data-driven algorithms for energy management systems: Use of reinforcement learning for energy management in households and businesses to optimize the demand response of flexible consumers, storage and generators.