Project > Overview
Redefining HPC, from infrastructure to application
We assembled a project group spanning a vast range of expertise, from liquid cooling to advanced programming, ensuring that we could take a holistic approach to design, implementation and testing. As well as a range of European HPC projects and partners, the group includes industrial SME focus, with MAX for FPGA data-flow, ICE for infrastructure, ARM for HPC tooling and ZPT to collapse the memory bottleneck – providing a rounded team well-equipped for the full solution stack.
The group worked in collaboration to co-design a balanced architecture for both compute and data-intensive applications, using a cost-efficient, modular-integration approach that harnesses novel inter-die links and taping out a EuroEXA processing unit that integrates FPGA for data-flow acceleration.
We provided a homogenised software platform that offers heterogeneous acceleration, with scalable shared memory access. We also created a unique, hybrid geographically addressed switching and topology interconnect within the rack, while adopting low-cost Ethernet switches – offering low-latency and high-switching bandwidth.
The project group also included a mix of experts in key HPC applications, from weather forecasting and astronomy, to life science and bioinformatics – offering us the opportunity to demonstrate the results of our work with the deployment of an integrated and operational PetaFLOPS level prototype hosted at STFC.
It’s a system that will manage local failures – with the support of run-to-completion platform-wide resilience mechanisms – while communicating with higher levels of the stack. As such, the EuroEXA project can monitor and control the platform’s advanced runtime capabilities – demonstrating a co-designed solution that supports both existing sub-ExaScale and project-developed ExaScale applications.