Cascade is commissioned
The fourth-generation system is up and running and produced its first weather forecast before its August 2025 public unveiling.
National compute · FULL DEEP TRACKER
NIWA Cascade Supercomputer
A living evidence file separating Cascade’s commissioned compute from Rapids’ archive capacity, and testing whether Crown-funded access translates into widely used, resilient and measurable national science and AI infrastructure.
OPERATIONAL · NATIONAL ACCESS TRACKINGHobsonville · Silverdale · AucklandCONTINUOUSLY MONITOREDUpdated 25 August 2026
COMPUTE2.4 petaflops
CPU CORES61,440
RAPIDS START22 PB
TOTAL INVESTMENTNZ$35m
WHY IT MATTERS
Strategic significance
Cascade and Rapids form nationally important digital research infrastructure for weather, climate, freshwater, oceans, hazards, remote sensing and AI. Unlike proposed data-centre capacity, the system is commissioned and producing forecasts. Its wider value depends on resilient operations, transparent science-sector access, sustained specialist capability and evidence that faster computation and larger data holdings improve decisions and research outcomes.
CURRENT ASSESSMENT
The infrastructure is operational; national utilisation and access outcomes are not yet transparent.
Cascade contains 61,440 CPU cores, 240TB of RAM and 7PB of local storage across 320 servers, delivering 2.4 petaflops. Rapids starts with 22PB of archive capacity and is designed to scale to 100PB over a decade. The two systems are located at separate CDC facilities in Hobsonville and Silverdale and linked by a high-speed network. Cascade has produced operational forecasts and materially improved NZCSM throughput.
The NZ$35 million headline combines compute and archive infrastructure and should not be treated as the price of Cascade alone. The strongest demonstrated outcome is the jump from 22 to 43 hours of NZCSM forecasting per compute hour. The next test is whether cross-sector access, utilisation, reliability, AI workloads and science outputs are measured publicly enough to establish national return on investment.
EVIDENCE GATES
What is confirmed—and what remains open.
2.4 petaflops of CPU compute
Published specifications include 61,440 CPU cores, 240TB RAM, 7PB storage and 320 servers; reviewed sources do not disclose a dedicated GPU count.
Rapids starts at 22PB
Rapids is a scientific data archive designed to scale to 100PB over ten years. Future capacity is not current installed storage.
Two CDC locations
Cascade and Rapids are distributed between Hobsonville and Silverdale with high-speed interconnection and disaster-recovery intent; exact workload failover arrangements are not public.
Forecast throughput nearly doubled
NZCSM testing improved from 22 to 43 forecast hours per compute hour, a concrete workload result rather than peak-spec marketing.
Crown-funded science-sector access
Earth Sciences New Zealand and REANNZ state the wider New Zealand science sector will receive access, but public eligibility, allocation and service-level rules are limited.
Energy savings forecast
Testing indicated about NZ$1.5m in energy savings over the system lifespan versus design plans; actual metered energy, PUE and carbon intensity are not disclosed.
Utilisation and outcome data missing
Queue time, utilisation, user institutions, project allocations, operating cost, uptime, incidents and quantified research outcomes are not publicly reported.
CAPACITY RECONCILIATION
Do not treat unlike numbers as equivalent.
Cascade peak performance
A compute-rate measure for the CPU system, not storage, AI model capability, utilisation or completed science.
Rapids starting capacity
Installed starting archive capacity distinct from Cascade’s 7PB local storage.
Rapids decade target
A future scalable archive design, not current commissioned capacity or funded data already stored.
DELIVERY & OPERATING STACK
The layers behind the headline.
ComputeCascade61,440 CPU cores and 240TB RAM run modelling, forecasting, simulation and additional AI workloads.
Fast storage7PBCascade-local storage supports active workloads and must not be added blindly to the separate archive headline.
Scientific archiveRapids · 22PBLong-term environmental data storage is designed to expand toward 100PB over a decade.
Data centres2 CDC sitesHobsonville and Silverdale provide physical security, power, cooling and geographic separation within Auckland.
ConnectivityHigh-speed inter-siteThe link lets systems communicate and supports access during public-network disruption; bandwidth and independent route detail are not public.
Access networkREANNZCrown-funded access connects the wider science sector, subject to allocation, identity, data and support controls.
OperationsEarth Sciences NZThe former NIWA team operates the facility after NIWA and GNS Science merged on 1 July 2025.
Vendor stackHPE + Xenon + partnersVAST Data, Versity, Spectra Logic and Altair support storage, archive and workload layers, creating lifecycle dependencies.
VERIFIED TIMELINE
Key milestones and status changes
- 2022–23NIWA begins multi-year planning and procurement for its fourth-generation supercomputing platform.
- 2024Construction, integration and migration planning progress while NIWA identifies Silverdale as the successor hosting location.
- 30 May 2025NIWA reports Cascade and Rapids are up and running at separate Hobsonville and Silverdale CDC data centres.
- 1 Jul 2025NIWA and GNS Science merge to form Earth Sciences New Zealand, transferring operational stewardship into the new organisation.
- 27 Aug 2025Cascade is publicly unveiled after producing its first forecast; the NZ$35m combined investment is announced.
- 2025 performanceNZCSM testing delivers 43 forecast hours per compute hour versus 22 on the previous system.
- CurrentCrown-funded access is stated for the national science sector; public allocation, utilisation and user-outcome reporting remain limited.
- Next decadeRapids is designed to scale from 22PB toward 100PB as environmental datasets and AI workloads grow.
STAKEHOLDER MAP
Who shapes the outcome
- Earth Sciences New Zealand and its supercomputing operations team
- Ministry of Business, Innovation and Employment and the Crown owner
- REANNZ and connected universities and research organisations
- CDC Data Centres and Auckland facility operations teams
- Hewlett Packard Enterprise, Xenon, VAST Data, Versity, Spectra Logic and Altair
- Weather, climate, hazard, ocean, freshwater, remote-sensing and AI researchers
- MetService integration teams and operational forecasting users
- Emergency management, infrastructure, primary-sector and public-policy decision makers
- Māori researchers, data kaitiaki and holders of environmental knowledge and datasets
- New Zealand taxpayers funding infrastructure and science access
DEPENDENCIES
What must align
- Reliable power, liquid cooling and physical operations at both CDC sites
- Diverse high-speed inter-site and REANNZ connectivity
- Specialist HPC, storage, modelling, cyber and user-support staff
- Software licensing, vendor support, spare parts and security patching
- Fair allocation between operational forecasting and contestable research workloads
- Data governance, privacy, sovereignty, classification and archive policies
- Stable operating funding across the expected six-year compute lifecycle
- User capability to optimise codes and workflows for the architecture
- A successor and expansion plan before compute and archive demand outgrow capacity
RISKS & OPEN QUESTIONS
What is not yet resolved
- Peak petaflops can be mistaken for sustained application performance or available user capacity
- The 100PB archive target can be misreported as current storage
- Crown-funded access does not guarantee open, immediate or equitable access for every researcher
- Both locations are in the Auckland region and may share regional power, network or hazard dependencies
- Renewable electricity claims describe supply or matching and do not disclose hourly marginal emissions
- Closed-loop cooling minimises operational water use but site-level PUE, WUE and total energy are not published
- Vendor and software dependencies can increase lifecycle cost or constrain future upgrades
- Rapid growth in AI and environmental data can exhaust compute, storage and specialist support before replacement
- Operational forecasting may appropriately displace external research during emergencies, reducing predictable access
- Without utilisation and outcome reporting, public return on the NZ$35m investment is difficult to assess
CORE EVIDENCE REGISTER
Primary and supporting sources
01Earth Sciences NZ: NZ$35m Cascade unveiling↗02NIWA: Cascade specifications, Rapids and access↗03Earth Sciences NZ: High Performance Computing Facility↗04NIWA Water & Atmosphere: Cascade technical report↗05CDC: Auckland facilities and operating capacity↗06CDC: Sustainability and renewable-electricity reporting↗07NIWA Statement of Corporate Intent 2024/25↗