Bede, the N8’s new high performance computing platform, is now open for users

By

Bede is the N8’s Power and GPU-based high-performance computing (HPC) platform.

Unlike traditional x86 platforms, Bede is GPU-based and makes use of NVIDIA’s NV Link high-bandwidth interconnects to move tensor outputs between the GPU and system memory. This unique architecture is ideally suited to artificial intelligence (AI) and machine learning (ML) applications. Bede will also be able to work with much higher resolution imagery than has previously been possible.

Since the system was handed over on 17 August, the team at Durham University have been working to fully commission the system ready for its formal launch. They have tailored the hardware and software environments to make it as easy as possible for researchers to maximise the benefits of the system’s unique GPU-based architecture.

Researchers who wish to use the system should visit the N8 CIR website for details of the application process: https://n8cir.org.uk/supporting-research/facilities/nice/bede-docs/bede_registrations/

Experienced users of high-performance platforms will be familiar with accessing a system such as Bede. However, researchers new to HPC shouldn’t be put off from applying for access. Each of the N8 universities has a research software engineer (RSE) dedicated to supporting Bede and its users, which can be found here: https://n8cir.org.uk/supporting-research/facilities/nice/rse-support-bede/

The system is not restricted to N8 CIR’s current research priorities of Digital Health and Digital Humanities. Researchers from all academic domains should consider applying for access. Each of the N8 Universities have equal access to the machine. In addition, there is a separate allocation for non-N8 EPSRC-funded researchers.

Full details of the system can be found on the N8CIR website.

The page includes overviews of the system’s hardware, its software environments, details of the comprehensive RSE support available to users and an explanation of how to register projects and user accounts.