UKRI National Compute Resources (NCRs)

Summary

UKRI has funded six National Compute Resources (NCRs) under phase one: Baskerville, Charger, Cirrus, Isambard 3, Mary Coombs and Zenith. Four are new or expanded systems announced in February 2026 for £76 million (Cirrus, Charger, Baskerville NCR and Cambridge's Zenith). Isambard 3 and Mary Coombs are existing systems that UKRI folded into the NCR portfolio from April 2026.

Three are CPU-based (Cirrus, Charger, Isambard 3) and three are GPU-based (Baskerville NCR, Mary Coombs, Zenith). NCRs aim to give researchers a varied, easier-to-use national compute landscape, including people who have never used a supercomputer. UKRI funding covers hardware and five years of service, up to 2031.

Background

NCRs are the first major step in the UK Compute Roadmap, a national plan published by the Government in July 2025. They sit alongside UKRI's flagship investments: the Next National Supercomputing Service (to be hosted at Edinburgh) and the AI Research Resource (AIRR). UKRI describes the NCR model as an evolution of its earlier Tier 2 approach, and says NCRs give continuity for users moving off Tier 2 services and ARCHER2. ARCHER2 access ends at 17:00 GMT on 20 November 2026.

Key features of the programme:

  • Diverse technology. Different hardware for different needs: CPUs for traditional simulation and many-task workloads, GPUs for AI and data-intensive work.
  • Easier access. Simplified systems and user support aimed at newcomers to high-performance computing.
  • Long-term support. Funding covers equipment plus five years of service to 2031.
  • UKRI-controlled access. UKRI keeps responsibility for the access policy, so most computing time is allocated under UKRI's policy rather than by the host university. NCR-specific access routes had not been published at the time the ARCHER2 migration pages were written.
  • Training and skills. UKRI has opened a funding call for Community Centres of Excellence, a service layer between users and the NCRs.

Architecture and size comparison

NCR Host Processor type Main compute Size Vendor and interconnect
Cirrus University of Edinburgh (EPCC) CPU only (x86) AMD EPYC 9825 “Turin”, two 144-core processors per node 640 nodes, 184,320 cores (after the August 2026 expansion) HPE Cray EX4000, Slingshot 11
Charger UCL (hosted at DataVita, Scotland) CPU (x86) Intel Xeon 6980P, 128 cores per processor (per a Cambridge talk) More than 37,000 cores HPE, Slingshot networking, Cray storage
Isambard 3 University of Bristol (BriCS, with the GW4 universities) CPU only (Arm) NVIDIA Grace CPU Superchip, two 72-core Neoverse V2 processors per node 384 nodes, 55,296 cores; about 2 PFLOP/s (HPL), about 300 kW HPE Cray XD224, Slingshot 11, 2 PB Lustre storage
Baskerville NCR University of Birmingham GPU NVIDIA B200 (per a Cambridge talk) Not yet published Lenovo, liquid-cooled (per trade press)
Mary Coombs STFC Hartree Centre, Daresbury GPU NVIDIA H100 157 nodes, 628 GPUs; 24.41 PFLOP/s InfiniBand NDR200
Zenith University of Cambridge GPU AMD Instinct MI355X, eight per node, plus two 64-core AMD EPYC 9575F processors 72 nodes, 576 GPUs in total; the NCR share is listed as 28 nodes, 224 GPUs Dell PowerEdge XE9785L, 400 Gb/s interconnect

This is the latest information we could find (01/11/2026) regarding the sizes and is subject to change.

Sources and links