National HPC (Tier-1) and Tier-2 Support
Overview
If your computational requirements have grown beyond what on-site University platforms can provide, then the UK Tier-1 and Tier-2 platforms may offer a solution. We aim to help you:
- apply for time on these platforms;
- run jobs and optimise performance of your code too.
For more information about the platforms, please see the sections below.
If you would like advice or help regarding Tier-1 and/or Tier-2, please contact us at:
National Service: Archer (Tier-1)
The Archer platform (based at Edinburgh) comprises 4920 compute nodes each with 24 CPU cores, i.e., a total of 118,080 cores. All nodes are connected by Cray low-latency/high-bandwidth Aries interconnect. Archer will be replaced by Archer2 so there are now no further calls for projects on Archer. However, Archer is still open for COVID research.
Archer/Archer2:
- is ideal for running large, parallel (i.e., true HPC) computational jobs
- it is not suitable for single-node high-memory work
Please see the Archer2 access page for how time on Archer2, once in service, can be applied for.
N8-CIR (formerly N8-HPC)
The original N8-HPC platform, Polaris, has now been decommissioned. A new GPU cluster, Bede, containing 128 Nvidia V100 GPUs and IBM Power9 CPUs for multi-node GPU capable applications is now available! Please see our Bede page for more details.
Tier-2 Platforms (incl. JADE)
There are six EPSRC-funded Tier-2
The centres are:
- CSD3 facility (led by University of Cambridge)
- The Materials Modelling Hub (led by University College London)
- JADE the Joint Academic Data Science Endeavour (led by University of Oxford)
- HPC Midlands Plus (led by Loughborough University)
- Isambard the GW4 Tier 2 HPC service (led by University of Bristol)
- Cirrus (led by University of Edinburgh)
See also http://www.hpc-uk.ac.uk/access/ and possibilities of DiRAC and maybe SCARF for those with the “STFC Remit”.
Manchester researchers may be particularly interested in JADE which comprises 22 * Nvidia DGX-1 deep learning systems, each of which contains eight Nvidia P100 GPUs connected by high-speed NV-link.