EuroEXA reports European Exascale Programme innovation

At launch one of the largest projects ever funded by EU Horizon 2020, EuroEXA aimed to develop technologies to meet the demands of Exascale high-performance computing (HPC) requirements and provide a ground-breaking platform for breakthrough processor-intensive applications.

EuroEXA brought together expertise from a range of disciplines across Europe, from leading technologists to end-user companies and academic organisations to design a solution capable of scaling to peak performance of 400 PetaFLOPS, with a peak power system envelope of 30 MW that approaches PUE parity using renewables and chassis-level precision liquid cooling.

Dr. Georgios Goumas EuroEXA Project Coordinator said, “Today, High-Performance Computing is ubiquitous and touches every aspect of human life. The need for massively scalable platforms to support AI-driven technology are critical to facilitate advances in every sector, from enabling more predictive medical diagnoses, treatment and outcomes to providing more accurate weather modelling so that, e.g. agriculture, can manage the effects of climate change on food production.”

EuroEXA Demonstrates EU Innovation on an equal footing with RoW

Meeting the need for a platform that answers the call for increased sustainability and lower operational carbon footprint, the 16-partner strong coalition delivered an energy-efficient solution. To do so, the partners overcame challenges throughout the development stack including energy efficiency, resilience, performance and scalability, programmability, and practicality.

The resulting innovations enable a more compact and cooler system, reducing both the cost per PetaFLOPS and its environmental impact; is robust and resilient across every component and manages faults without extended downtime; provides a manageable platform that will continue to provide Exascale performance as it grows in size and complexity; harnesses open-source systems to ensure the widest possible range of applications, ensuring it is relevant and able to impact real-world applications.

The project extended and matured leading European software stack components and productive programming model support for FPGA and exascale platforms, with advances in Maxeler MaxJ, OmpSs, GPI and BeeGFS. It built expertise in state-of-the-art FPGA programming through the porting and optimisation of thirteen FPGA-accelerated applications, in the HPC domains of Climate and Weather, Physics and Energy, and Life Sciences and Bioinformatics, at multiple centres across Europe.

EuroEXA Innovation Being Applied Today at ECMWF and Neurasmus

A prototype of a weather prediction model component extracted from ECMWF’s IFS suite demonstrated significantly better energy-to-solution than on current HPC nodes, achieving a 3x improvement over an optimized GPU version running on an NVIDIA Volta GPU. Such an improvement in execution efficiency provides an exciting avenue for more power-efficient weather prediction in the future.

Further successful outcomes were made through the healthcare partnership with the Neurasmus programme at the Amsterdam UMC, where brain activity is being investigated. The platform was used to generate more accurate neuron simulations than has previously been possible, helping to predict more accurately healthcare outcomes for patients.

EuroEXA Legacy – Extensive FPGA Testbed an Aid to Further Developments

Outcomes generated include deploying what is believed to be the world’s largest network cluster of FPGA (Field Programmable Gate Array) testbeds, configured to drive high-speed multi-protocol interconnect, with Ethernet switches providing low-latency and high-switching bandwidth.

The original proposal was for three FPGA clusters across the European partnership. However, Covid-19 travel restrictions necessitated an increased resource of 22 testbeds, developed in various partner locations. This has benefited the project by accelerating through the massively increased permutations and iterations available, which has also provided a blueprint for several partners to develop high-performance FPGA-based technologies.