Inspection: Guiding Optimizations with Performance and Energy Models
This work package aims to implement the inspection phase of the SparCity framework by developing performance, energy, and machine-learning-based prediction models. This WP is essential to make intelligent node-level and system-level optimizations and co-designing future hardware and applications.
INESC-ID (Lead partner), KU, Simula, LMU, SU, Graphcore
Node-Level Static and Dynamic Optimizations
The main aim of this work package is to develop node-level static and dynamic optimizations designed specifically for sparse computations. This WP has the objective of automating many non-trivial data locality optimizations as well as develop custom mixed-precision solutions for applications that use graphs or sparse tensors.
KU (Lead partner), SU, Simula, LMU, INESC-ID, Graphcore
System-Level Static and Dynamic Optimizations
This work package develops system-level static and dynamic optimizations for sparse computations. Its main objective is to reduce communication overhead and overlap communication with computation.
LMU (Lead partner), SU, KU, Simula, INESC-ID, Graphcore
Co-design: Digital Twin for HPC
The main aim of this work package is to develop hardware inspection tools and a digital twin of a supercomputer for co-design purposes.
SU (Lead partner), INESC-ID, KU, Simula, LMU, Graphcore
Co-design: Demonstration with Real-Life Applications
This work package aims to enhance the performance and energy efficiency of four real-world applications.
Simula (Lead partner), INESC-ID, KU, LMU, SU, Graphcore
Project, Innovation, and Risk Management
This work package is related to the effective and timely implementation of the project, the quality control of the results, the risk management of the project as a whole, as well as a timely and necessary interaction with the European Commission and other interested parties.
KU (Lead partner), SU, Simula, INESC-ID, LMU, Graphcore
Communication, Dissemination and Exploitation
This work package defines the communication, dissemination and exploitation strategies to maximize the impact of the project’s results. This will be achieved through divulgation activities to the stakeholders and other parties, the establishment of relationships with relevant projects and initiatives and publications in highly ranked scientific journals, conference proceedings and books.
INESC-ID (Lead partner), KU, LMU, Simula, SU, Simula, Graphcore