{"resourceId":"leeds-calder-hpc-first-week-156k-jobs-90-percent-gpu-utilisation-20261021","versions":[{"version":"external-aa616ba51338c25e2c33bde69ac841e89c5953b9f4bd6be37a43c3a297d09360","resource":{"id":"leeds-calder-hpc-first-week-156k-jobs-90-percent-gpu-utilisation-20261021","title":"University of Leeds Calder HPC first week (21 Oct): over 156,000 jobs, 90.3% average GPU utilisation on NVIDIA H200 NVL nodes and 82.2% peak utilisation, while proprietary applications and Slurm tuning are still catching up","organization":"University of Leeds Research Computing (Advanced Research Computing)","sector":"University research computing","geography":"Leeds, United Kingdom (transferable to U.S. university research computing centres)","publishedAt":"21 October 2026 (ARC Leeds news post)","publicationDate":"2026-10-21","eventDate":"2026-10-09","sourceName":"University of Leeds Advanced Research Computing","sourceLabel":"ARC Leeds, \"Calder: a strong start for our new HPC\", Patricia Ternes and Jo Kershaw, 21 October 2026","sourceUrl":"https://arc.leeds.ac.uk/knowledge-centre/calder-hpc-first-week/","evidenceClass":"vendor-claim","outcomeClass":"emerging","topics":["infrastructure","operating-model"],"finding":"The University of Leeds research computing team reports that in Calder's first seven days of service, researchers submitted over 156,000 jobs across a range of disciplines; GPU utilisation averaged 90.3% over the week and thousands of CPU jobs drove peak utilisation to 82.2%. Two researchers describe early use: a 3D scene-understanding group submitted over 2,700 CPU and 500 GPU jobs and said the H200 GPUs' large memory let them run big 3D models on whole scenes, and an Earth-sciences postdoc said the H200 NVL GPUs considerably sped up training of physics-informed neural networks. The team says the familiar setup carried over from the existing Aire cluster, but several applications, particularly proprietary software, are not yet available on Calder, users depending on them are asked to stay on Aire, and the team is still refining Slurm for growing and diverse workloads. The 9 October launch post says Calder was procured through an academic-led process approved by the Research IT Steering Group (MTI with Dell hardware, support from Cambridge Research Computing), with security checks and user testing before launch.","sledRelevance":"A rare early operating metric for a new university AI and HPC cluster. It shows demand for GPU capacity can saturate a new system within a week, and that software availability and scheduler tuning lag hardware, useful benchmarks for U.S. research computing centres planning launches and capacity.","evidence":"Read the full 21 Oct 2026 ARC Leeds post (datePublished in page metadata): 156,000+ jobs in seven days, 90.3% average GPU utilisation, 82.2% peak utilisation, two named researcher testimonials, proprietary software gap and Slurm refinement. Opened the 9 Oct 2026 launch post for procurement (Research IT Steering Group, MTI, Dell, Cambridge Research Computing) and pre-launch security checks and user testing.","architectureImplications":"Hybrid CPU and NVIDIA H200 NVL GPU cluster run alongside an existing system (Aire) under Slurm; keeping the user environment consistent with the previous cluster eased migration.","governanceImplications":"Academic-led procurement via a Research IT Steering Group; capacity allocation and fair-share policies will matter as GPU demand stays near saturation.","securityPrivacyImplications":"Security checks were completed before launch; no further detail is given.","caveats":"Self-reported by the operating team in a promotional post; no definitions of utilisation, queue wait data, user counts, GPU counts or baseline comparison with Aire are given. One week of data, likely affected by launch-period backlog migration. Testimonials are selected by the operator. GPU model and counts beyond \"H200 NVL\" are not stated in the inspected posts.","streamIds":["research"],"roles":{"sales":"Lead with launch readiness, not just hardware. Leeds reports its new Calder cluster ran over 156,000 jobs and averaged 90.3% GPU utilisation in its first week, yet proprietary applications were not yet available and Slurm was still being tuned. Research computing directors and CIOs planning GPU expansions face the same gap. Interpretation: a bounded launch-readiness or capacity-planning engagement fits; one week of operator-reported data cannot support claims about sustained utilisation, research outcomes or return on investment.","engineering":"Frame the fit around migration and scheduling. Calder runs CPU and H200 NVL GPU partitions under Slurm alongside the existing Aire cluster, with a familiar environment that let users be productive from day one. Pre-sales should confirm software licensing readiness, fair-share and partition design for saturated GPU demand, and how utilisation will be measured. Interpretation: plan scheduler policy and application porting before launch, because the source shows both still in progress after go-live.","delivery":"Treat delivery as a staged migration with measurement. Work covers porting and licensing applications, configuring Slurm partitions and fair-share, running parallel service with the legacy cluster, and publishing utilisation and queue metrics. Research computing owns the service under the Research IT Steering Group. Interpretation: acceptance should include defined utilisation and queue-wait metrics and application availability targets, since the source reports headline utilisation without definitions or wait data."},"roleDetails":{"sales":{"customerProblem":"Universities launching new GPU clusters face immediate demand saturation while software availability and scheduling policy lag.","stakeholders":["Research computing director","CIO","Research IT steering group","Heavy GPU research groups"],"discoveryQuestions":["Which applications must be ready on day one?","How will GPU time be allocated when demand exceeds supply?","What utilisation and wait metrics will you report?"],"valueHypothesis":"Launch-readiness planning may shorten the period in which users must stay on legacy systems and make GPU allocation transparent.","potentialEngagement":"A launch-readiness review covering application porting, licensing, scheduler policy and metrics for one planned cluster.","unsupportedClaims":["Sustained 90% GPU utilisation","Research productivity gains","Return on investment"]},"engineering":{"fit":"Fits universities adding GPU capacity alongside an existing HPC system.","architectureIntegration":"CPU and H200 NVL GPU partitions under Slurm, run in parallel with a legacy cluster and a consistent user environment.","prerequisites":["Application porting and licences","Scheduler and fair-share design","Monitoring of utilisation and waits"],"constraints":["One week of operator data","No queue or user metrics","Proprietary software lag"],"security":"Complete security checks before launch and keep consistent access controls across old and new clusters.","proofOfValue":["Define and report utilisation and queue waits","Track application availability against plan","Compare job throughput with legacy cluster"]},"delivery":{"work":"Port and license applications, configure scheduling, run parallel service, collect metrics and gather user feedback.","dependencies":["Software vendor licences","Integrator support","Researcher user testing"],"ownership":"Research computing service, governed by a research IT steering group.","skillsAdoption":"Keep the user environment consistent to limit retraining; train staff on scheduler tuning for mixed workloads.","governanceCheckpoints":["Pre-launch security and user testing sign-off","Scheduler policy review after first month","Legacy cluster retirement decision"],"acceptanceCriteria":["Proposed: priority applications available by an agreed date","Proposed: published queue-wait metric by partition","Proposed: defined utilisation measure reported monthly"],"risks":["GPU saturation causing long waits","Licensing delays","Unclear metrics overstating success"]}},"retrievedAt":"2026-10-25T22:05:16-05:00","enrichedAt":"2026-10-25T22:05:17-05:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Familiar environment reduced onboarding effort; behind-the-scenes infrastructure staff effort is emphasised but not quantified.","procurementImplications":"Academic-led tender with an integrator and external support partner; software licensing (proprietary applications) lagged hardware readiness.","operatingModelImplications":"Running new and legacy clusters in parallel during migration, with community feedback channels and continuous scheduler tuning, is the stated operating approach.","sourceVerification":{"openedUrl":"https://arc.leeds.ac.uk/knowledge-centre/calder-hpc-first-week/","referenceExcerpt":"In its first week, researchers submitted over 156,000 jobs, with GPU utilisation averaging 90.3%, already supporting impactful research across the University.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}