Bede is the N8’s newest high performance computing (HPC) platform. Since its formal launch in December 2020 researchers have been using the system across a range of academic domains to accelerate their research.
About Bede
Bede is classed as a Tier 2 supercomputer and so falls between many universities’ on-premise HPC clusters but below Tier 1 clusters such as ARCHER. Bede was funded by the Engineering and Physical Sciences Council (EPSRC) as part of a programme to evaluate new HPC platforms and architectures. It will provide a vital step for researchers from local to national clusters.
Unlike many traditional X86 supercomputers, Bede makes use of IBM’s POWER9 CPU coupled to NVIDIA GPUs. It uses NVIDIA’s NVLink to efficiently move the outputs from calculations from the GPU to the larger system memory. This architecture is ideal for machine learning (ML) and artificial intelligence (AI) work. Bede will also be able to work with much higher resolution imagery than has previously been possible.
RSE Support
N8 CIR recognise that Bede’s unique architecture may be a little daunting, even to the more experienced HPC user. To help address this concern each of the N8 sites has a research software engineer (RSE) dedicated to supporting projects on Bede.
Working closely with an RSE will be vital to unlocking Bede's potential. Marion Weinzierl, N8 CIR's research software engineering theme leader, will be at the event and will be able to answer general questions about the available support.
Case Studies
During this virtual event we will also hear from some of the researchers that have already used Bede in their work. They will speak about their previous experience of HPC platforms and then discuss the challenges of modifying code to maximise the benefits of Bede’s GPU-based architecture. From there they will talk about how the system has helped accelerate their research.
Marta Camps Santasmasas
PDRA, The University of Manchester
Turbulent fluid flows with GASCANS
Fluid flow around us and our environment affects our day to day lives in numerous ways; from the ventilation in our homes and the wind down our street to the aerodynamics of an aeroplane. Airflow in these cases is generally turbulent, defined by fluctuations over a large range of temporal and spatial scales. Turbulent flows are simulated using Computational fluid dynamics (CFD) codes, which typically need thousands of CPU cores to account for these scales and obtain accurate results.
GASCANS is a CFD code for turbulent flow based on the lattice Boltzmann method, and runs on GPU hardware. Its aim is to reduce the computational resources needed to simulate turbulence. We have already tested GASCANS for wind around a single building and we aim to use BEDE to simulate larger and more complex urban environments as well as exploring its application to several other sectors.
Thomas Hudson
PGR, Durham University
Analysing Text Using Machine Learning
Our project involves using machine learning to analyse long text documents such as books, movie scripts, and news articles.
Current techniques for common tasks such as translation and document summarization only perform well on a few short sentences or paragraphs. We aim to use BEDE to enable processing documents of tens or thousands of words, e.g., translating an entire book from French to English, detecting news articles with bias, identifying plagiarised sections in long pieces of work, and summarising academic papers alongside many other applications.
Paul Richmond
Senior Lecturer/Research Software Engineering Fellow, University of Sheffield
Flame GPU
Complex Systems exist all around us from systems of cells in the body to systems of migratory behaviour and even the economy. Complex systems can be studied with a technique called agent based modelling in which individuals are simulated with the system level properties emerging as a result of interactions.
This process is computationally expensive. FLAME GPU2 is a piece of software which is able to use the processing power of modern GPU hardware to perform large scale simulations of complex systems. It does so by creating a simple abstraction layer which hides the complexity of the underlying hardware.
This case study will present the FLAME GPU software and its use cases.
Holger Schulz
PDRA, Durham University
PDE-solving with 3 Levels of Parallelism
We intend to utilize Bede for the project Peano/ExaHype2 which provides an engine
for the solving of hyperbolic partial differential equations. We are particularly interested
in the interplay of MPI, OpenMP and offloading to the GPU in a task-based approach.
The technology developed with Bede will hopefully serve as a demonstrator for achieving at
least partial Exascale computing."
More Information
You can find out more about Bede, including the hardware, software, support and application process on the N8 CIR website at: https://n8cir.org.uk/supporting-research/facilities/bede/.
As this event is taking place online, the joining links will be sent out via Eventbrite on Wednesday 27 January 2021.
You can register at: https://www.eventbrite.co.uk/e/bede-introductory-event-tickets-133064348047.
