top of page


  • Writer's pictureRefMap

Auto-tuning Multi-GPU High-Fidelity Numerical Simulations for Urban Air Mobility

A new research paper from RefMap partners was accepted for publication in the DATE conference, the prestigious Design, Automation, and Test in Europe Conference. The title is “Auto-tuning Multi-GPU High-Fidelity Numerical Simulations for Urban Air Mobility”. The paper is the result of the joined research efforts and close collaboration between RefMap partners Institute of Communication and Computer Systems (ICCS) and Kungliga Tekniska Hoegskolan (KTH).

Namely, Sotiris Xydis from our partner ICCS will present the above research work by showcasing the first promising results of RefMap's strategy to optimise high-fidelity numerical simulations for UAM (Urban Air Mobility) using GPU acceleration capabilities and auto-tuning techniques.


The aviation field is rapidly evolving towards an era where both typical aviation and Unmanned Aircraft Systems are essential and co-exist in the same airspace. This new territory raises important concerns regarding environmental impact, safety, and societal acceptance.


The RefMap project is an initiative that addresses these issues and aims at optimising air traffic in terms of the environmental footprint in aviation and drone flights. One of RefMap's objectives is the development of powerful deep-learning models that predict urban flow based on extensive CFD simulations. However, the excessive time requirements of CFD simulations require the computational power of exascale heterogeneous supercomputer clusters.


This work motivates the need for HPC (High-Performance Computing) optimisation of these LES simulations to acquire training data in a high-throughput fashion and presents the acceleration capabilities of powerful state-of-the-art GPU-enabled CFD solvers. Then, it presents RefMap's advanced optimisation approach that leverages a GPU-enabled solver with MPI support and extends it with an auto-tuning tool to create an optimization framework for portable and highly efficient simulation deployment on different parallel systems and GPU architectures.

You can find the paper here,



bottom of page