FOR 5880:  Holistic Energy and Performance Modeling for Sustainable Computing (Mod4Comp)

• Subject Area: Computer Science, Systems and Electrical Engineering
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 545776403

The exponential growth in global ICT energy consumption is making a significant contribution to global climate change. Data centers, mobile devices, the AI boom and autonomous driving are all playing their part and require a reduction in the power consumption of computers on a global scale. The aim of the research group "Holistic Energy and Performance Modelling for Sustainable Computing (Mod4Comp)" is to provide models for predicting the required computing time and energy demand in computing systems with the aim of supporting the energy-aware design of algorithms and new computer architectures. A holistic view of the energy and performance aspects requires precise knowledge of the interaction of hardware and software, broken down to the elementary components of the architecture: processors, caches, memory, data paths and communication networks. Such a view, with appropriate modelling to predict runtime and energy behaviour, has been lacking until now. Holistic also means that the modelling must not happen in isolation from the applications and that the models found must be verified by measuring them on the hardware of real architectures. To achieve this, the research group is pursuing the following approach: (i) Representative performance and energy models: For selected applications representing a variety of classes of real-world computational workloads, quantitative performance and energy models are built for a set of relevant architectures. The architectures include classical HPC computers with CPUs and GPUs, future embedded HPC as well as new neuromorphic architectures. (ii) Validation of models through advanced simulation and measurement: To make accurate quantitative predictions, novel simulation approaches will be developed and applied. The simulation integrates the power and energy models with sophisticated physical measurements that will also be developed. This will serve parameter fitting and validation of reliable models. (iii) Holistic modelling framework. The power and energy models will be implemented in a framework ranging from chip, node to network level and taking into account appropriate interfaces. This allows flexible creation of holistic models from interchangeable components. To cover a wide range of use cases, the framework combines several techniques such as analytical models, grey-box models and simulation on all levels of the hardware architecture. To achieve these goals, experts from computer architecture, electrical engineering, neuroinformatics, and parallel algorithm technology from FAU Erlangen-Nürnberg, TU Dresden and Forschungszentrum Jülich are working together in Mod4Comp.

• DFG Programme: Research Units

Projects

• Analytic chip-level performance, power, and energy modeling (Applicant Hager, Georg )
• Coordination Funds (Applicant Fey, Dietmar )
• Energy and performance modeling of the neuromorphic supercomputer SpiNNcloud Subproject SP6 within the proposed Research Unit "Holistic Energy and Performance Modeling for Sustainable Computing (Mod4Comp)" (Applicants Mayr, Christian ;  Partzsch, Johannes )
• Experimental node-level energy efficiency analysis (Applicants Nagel, Wolfgang E. ;  Schöne, Robert )
• Modeling of near-memory computing architectures for computer vision applications targeting performance and energy (Applicant Goehringer, Diana )
• MOdelling and Simulating In-memory Computing on chip and node level (MOSIC) (Applicant Fey, Dietmar )
• Performance and energy modeling for spiking network simulations on conventional and accelerator architectures (Applicant Kunkel, Susanne )
• Simulator framework for runtime and energy prediction of massively parallel message-passing programs (Applicant Wellein, Gerhard )

• Spokesperson: Professor Dr.-Ing. Dietmar Fey