Towards a more sustainable High Performance Computing sector: a hardware/software co-design proof-of-concept

The information and communication technologies (ICT) sector has become an inextricable part of our modern world, from supporting basic day-to-day tasks to performing complex simulations and calculations that are essential to scientific research, engineering, security, control and other fields. High-Performance Computing (HPC) systems that are capable of solving hugely complex and demanding problems with high computational power, are critical in the development of leading-edge global scientific and engineering projects currently underway in the commercial, academic and government spaces, spurring discoveries and innovations. With the rise of artificial intelligence and big data, it is inevitable that the percentage of global power consumption from HPC systems will grow rapidly and is going to be a major challenge in the future. As such, there is growing interest in how the expansion of HPC systems can be delivered in a manner that is environmentally sustainable and in line with the government's pledge for a net zero society.


This project will investigate how to exploit configurable and customisable processing technologies to design more sustainable HPC systems. The team of investigators will use a co-design approach, tuning processing hardware and software in collaboration to optimise HPC systems for both performance and energy efficiency. The proposed approach will be used to look at how to improve the performance and sustainability of wind energy systems, which in turn can be used to power HPC systems, further reducing the environmental impact of these systems. This is our primary research objective. For this exercise, we will dramatically reduce the environmental impact of Xcompact3d, an open-source software designed to study numerically fluid flows. It is currently among the most used softwares on ARCHER2, the UK supercomputing service. We will focus on Xcompact3d's wind farm simulator, a tool which can faithfully replicate virtually realistic scenarios encounter by modern wind farms. It will be re-designed with a hardware-software co-design approach. We will focus on three hardware architectures; 1) Field Programmable Gate Arrays (FPGAs), 2) Coarse Grained Reconfigurable Architectures (CGRAs), 3) RISC-V.

With FPGAs we are able to undertake the design of low-level bespoke memory access designs or algorithm level concurrency, with CGRAs to explore different approaches for mapping algorithms to the hardware and communication, and
RISC-V to experiment with different CPU and accelerator designs and their configurations. With support from the vendors, all these experiments will result in actionable results that will be use to further direct the research, but also for the vendors to understand how to best enhance their products for the HPC community.

The team of investigators will also gain insights into sustainability efforts in HPC with an in-depth study focusing on possible sustainability and net zero strategies. One of the overarching goals of the project is to provide a road-map for a more sustainable HPC landscape, and transfer the outcomes of the project to policy-making for a wide range of academic, government and industry stakeholders.

The project team is composed of experts from a variety of fields, including high performance computing, computer science, programming, computational fluid dynamics, innovation and sustainability, and policy-making. This interdisciplinary team will be able to bring a wide range of expertise to bear on the challenges of developing and utilising more energy and resource efficient HPC systems.