Work Package 2
AI- and HPC-Cross Methods at Exascale
This work package aims at building the AI and HPC at Exascale backbone of RAISE, providing the hardware and software infrastructure needed for the implementation of the compute- and data- driven use-cases of WP3 and WP4. HPC at Exascale is tackled in two ways:
the available production systems at the partners’ HPC centers together with appropriate documentation are made available and
new prototypes and disruptive technologies from the DEEP-projects and in the context of Quantum Computing are integrated in existing environments for testing, porting, and benchmarking purposes.
The operators of the systems support the application developers in the use of these platforms. All these activities in conjunction with requirement specifications of the use-cases provide co-design information, driving future developments of the hardware and software tools used.
To bundle the individual activities for the use-cases and to generate value for the wider community, it is furthermore the aim to develop a strategy and design plan creating a unique framework that in the future will hold trained Convolutional Neural Networks, algorithms for training, and which is optimized for application in Modular Supercomputing Archutecture environments. Therefore, corresponding interfaces are defined and the developments in the use-cases are monitored to meet the requirements of such a unique framework. The objective, which is the most crucial to the successful implementation of the use-cases is the development of cross- sectional AI methods, which are
utilizing heterogeneous/modular HPC architectures and scale efficiently, and are
designed and implemented in a generalized way to be applicable to a wide set of applications.
Tasks in WP2
Modular and heterogeneous supercomputing architectures
ParTec, UOI, RWTH, RTU
ParTec, UOI, RTU, BSC
Benchmarking on disruptive technologies
UOI, BULL, RTU
Software design of a unique AI framework
Cross-Sectional AI Methods
FZJ, CERN, BULL, RTU, FM