Hardware-aware algorithms in Scientific Computing

Funded by the German Academic Exchange Service (DAAD) as part of its programme indo-german partnerships (IGP) we were granted a joint project with the Department of Computational and Data Sciences (CDS) at Indian Institute of Science Bangalore (IISc).

Associated partners are the Tata Institute of Fundamental Research, Centre for Applicable Mathematics (TIFR-CAM), Bangalore, the Jülich Supercomputing Centre as part of the Forschungszentrum Jülich as well as NVIDIA India as an industrial partner.

Resposible PIs

Peter Bastian and Sashikumaar Ganesan.

Short description of the project

Algorithms in scientific computing, always eager for compute-power, need to adapt to the situation of ever increasing parallelism, corresponding structures and data locality. New algorithms and implementations need to be devised that are able to exploit and adapt to the new hardware architectures available in the future, despite the slowdown in Moore’s law. The main objective of this initiative is to bring together research activities at the Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University and Department of Computational and Data Sciences (CDS), Indian Institute of Science Bangalore (IISc) and at the associate partners, Jülich Supercomputing Centre (JSC) at Forschungszentrum Jülich (FZJ), Tata Institute for Fundamental Research (TIFR-CAM), Bangalore, NVidia, India, on hardware-aware scientific computing on future heterogeneous supercomputers, to develop young researchers and to initiate long term collaborative projects on this challenging field. The research at the lead institutions CDS and IWR as well as the associate partners FZJ, Tata Institute and industrial partner NVIDIA fits ideally together. There is a core overlap in specific fields such as finite element methods and solvers and there is complementary expertise, e.g. in uncertainty quantification, machine learning and computer architecture. The involved scientists are well-known in their respective fields. Moreover, the involvement of the industrial partner NVIDIA (India) is highly relevant to the topic of this program.

The key objectives are:

  • to train a new generation of students on master and Ph.D. level, who are both fluent in state-of-the-art efficient numerical methods and hardware-aware design and implementation of these algorithms,
  • to build up a leading network of researchers in India and Germany in high- performance scientific computing,
  • to establish collaboration on joint research projects among the project partners, and in the long term perspective,
  • to establish a cotutelle Ph.D. program between Heidelberg University and IISc along the lines of the existing IWR-DBT program which is concentrating on Big Data Research in the biosciences.

To attain these goals, the considered field is structured into five research themes

  1. Efficient numerical discretization schemes for PDEs
  2. Scalable and robust algebraic solvers
  3. Uncertainty quantification (UQ), machine learning (ML) and inverse problems (IP)
  4. Co-design of PDE applications for exascale architectures and technologies
  5. Software development for PDE-based methods
Impressum  |  Haftungsausschluss  |  Datenschutzerklärung  |  generated with Hugo v0.69.0