RSE Personal Research Fellowship awarded to EPCC's Nick Brown

Decarbonising our supercomputers is crucial, not only to ensure we meet our climate goals, but because drawing less energy also means that these systems will be cheaper and easier to install and operate. To this end, the question of how novel, more energy efficient, hardware can be exploited is of paramount interest and will be the focus of my research.  

One aspect of the Research Fellowship will centre on in-situ HPC data analytics. Large-scale simulation codes, such as models from the Met Office, generate vast amounts of raw data that must then be post-processed into information so that it is consumable by users. 

The state of the art is to use CPUs, however this wastes resource that could be otherwise used for compute and, furthermore, requires significant energy. My hypothesis is that Field Programmable Gate Arrays (FPGAs) would be a much better match and I think they can deliver a step change in capability. This is because by combining the very high bandwidth of FPGAs with their inherent energy efficiency, we can put them in the data plane (the part of a network responsible for the movement of data) if they reside at the network level. Consequently, FPGAs could undertake post-processing of vast amounts of raw data in real time as it is streamed out from simulation codes. 

My other area of interest is exploring the computational acceleration of the Met Office’s CASIM microphysics model on AMD Xilinx AI Engines (AIEs). These are provided in arrays of 400 engines, each running at 1.2GHz and capable of undertaking 8 floating point operations per cycle – again at very low energy.  

While the AIEs deliver significant raw compute power at low energy usage, the key challenge is how they might actually be leveraged in a code! Consequently, using the CASIM microphysics model as a vehicle to explore the potential advantages of this new technology will not only help develop performance best practice but I hope it will also deliver benefits to this production code as the demands on it, driven by scientific ambition, continue to increase. 

I am looking forward to starting my Personal Research Fellowship in 2024. It has in part been made possible by EPCC’s involvement in the Met Office Academic Partnership (MOAP) programme. Not only has MOAP opened up access to Met Office simulation codes, enabling me to talk to Met Office staff and understand how innovative research might help tackle some of their challenges, but the MOAP collaborative programme means that I will be able to ensure that the outcomes of my RSE Personal Research Fellowship can be fed back to them. 

One of the benefits of working at EPCC is that I am able to run my workloads on a wealth of HPC machines. The ExCALIBUR H&ES FPGA testbed, which contains a variety of networked FPGAs, will be the one I use most frequently but it will also be very helpful to have access to more traditional systems such as ARCHER2 and Cirrus so that I can compare and contrast my new approaches against the current state of the art.

Links 

RSE Personal Research Fellowships 
https://rse.org.uk/funding-collaboration/award/rse-personal-research-fellowships/ 

Met Office Academic Partnership at EPCC 
https://www.epcc.ed.ac.uk/whats-happening/articles/met-office-academic-partnership