Research Infrastructure

Bede – N8-CIR GPU Cluster

Local information about N8 CIR‘s GPU cluster, named bede.

A new GPU Cluster – Bede

N8 CIR’s recently-announced supercomputer, Bede, comprises of 32 x IBM Power9 dual-CPU nodes, each with 4 NVIDIA Volta V100 GPUs, giving a total of 128 GPUs. Connectivity within a node is optimised by both the CPUs and GPUs being connected to an NVIDIA NVLink 2.0 bus, and outside of a node by a dual-rail Mellanox EDR InfiniBand interconnect allowing GPUDirect RDMA communications (direct memory transfers to/from GPU memory).

A comprehensive overview of Bede’s hardware, including details of the login nodes, GPU nodes and inference nodes can be found at the N8CIR Bede hardware page.

Requesting access

Users running multi-GPU codes that can scale up to multi-node usage, or users who wish to develop and test such codes can request access to bede by following the instructions at the N8-CIR Bede project registration page. This asks for some brief details about your project. Access to Bede is on a per-project basis so that jobs run on the system are accounted in the correct project. The project PI should make the project registration and then login accounts on bede can be set up for anyone working on that project.

If you would like to discuss your application before submitting it, please email manchester-bede-support@n8cir.org.uk and we’ll be in touch.

Requesting support

The Research Infrastructure Team here in Manchester have some dedicated resource to help with all Bede related queries. Please email manchester-bede-support@n8cir.org.uk (including as much info as possible to help us diagnose any problems).

Please also see the N8CIR Bede website for access to online documentation, lists of installed apps and access to RSE support.

PI Project Management

PIs with projects on Bede can manage their projects (user access) via SAFE (EPCC’s tier1&2 supercomputer project management site).

Last modified on March 16, 2021 at 2:38 pm by George Leaver