The UK’s Science and Technology Facilities Council signed an agreement with Lenovo to install a new supercomputer at the STFC Hartree Centre. The GPU-based system will deliver 44.7 petaflops, more than 44 quadrillion calculations per second, and provide ten times the processing power of the centre’s existing Scafell Pike machine. Despite the increase, it is expected to use less electricity per unit of performance.
The installation forms part of the £210 million Hartree National Centre for Digital Innovation programme, which gives UK industry access to advanced technology and expertise. It will be housed in a new £30 million supercomputing centre at STFC’s Daresbury Laboratory in the Liverpool City Region. The investment links physical infrastructure with a programme intended to help businesses adopt supercomputing, AI and other emerging technologies.
GPU Compute Expands the Centre’s AI Capacity
The Lenovo ThinkSystem Neptune is designed around GPUs, which can perform many calculations in parallel and are well suited to contemporary AI workloads. Its 44.7-petaflop performance marks a substantial change in what the Hartree Centre can offer industrial partners. Larger models, more detailed simulations and more extensive datasets can be processed within practical timescales.
Raw capacity is only useful when organisations can translate it into projects. The Hartree Centre is the UK’s only supercomputing centre dedicated to industry engagement. This mission gives the system a different role from a machine focused solely on academic rankings. Its value will be measured by whether companies can use the compute to test ideas, improve products and develop technical skills they would struggle to acquire independently.
Warm-Water Cooling Reduces the Energy Penalty
Lenovo’s direct water-cooling technology is central to the design. The system circulates warm water close to heat-generating components, removing energy more effectively than conventional air cooling. The source reported that the approach can reduce energy demand by up to 40% while improving performance by as much as 10%. It also allows the system to occupy less physical space.
This efficiency matters because increasing AI compute can otherwise create a parallel increase in electricity and cooling requirements. A system ten times more powerful than its predecessor would be difficult to justify if energy use rose at the same rate. Direct cooling does not eliminate environmental impact, but it improves the amount of useful computation obtained from the power consumed.
Public Infrastructure Can Lower the Adoption Barrier
Many businesses cannot purchase and operate a supercomputer of this class. Hardware costs are only part of the barrier; specialised facilities, cooling, software and staff are also required. The HNCDI programme creates shared access to these resources, allowing companies to explore workloads before deciding whether and how to build their own long-term capability.
The programme also provides skills and technical knowledge. This human support is essential because moving an ordinary workload onto GPUs does not automatically make it efficient. Teams need to prepare data, select software and understand how their problem maps to parallel hardware. Hartree expertise can help industrial users avoid treating supercomputing as a simple rental of processing time.
The Facility Connects Several UK Technology Investments
The new machine complements the wider AI Research Resource and sits alongside HNCDI work in AI, quantum computing and supercomputing. These programmes serve different users and stages of innovation, but together they strengthen a national infrastructure base. The Daresbury facility gives that strategy a physical centre where equipment and expertise can be brought together.
Installation at the new building was planned for later in 2024. Delivery will involve more than connecting hardware. The centre must validate performance, integrate the system with existing services and create reliable access for partners. Security and scheduling will also matter when multiple organisations use shared infrastructure and bring commercially sensitive data or code.
The agreement demonstrates that AI capacity is increasingly an engineering question. Model capability depends on processors, interconnects, cooling, power and facilities as well as algorithms. The Hartree project addresses these layers together: GPU compute provides speed, water cooling improves efficiency and the industrial programme supplies access and knowledge.
Its strongest promise is not the number of calculations alone. It is the possibility that UK companies can use advanced infrastructure without first becoming supercomputer operators. If the centre converts its tenfold performance increase into successful industrial work, the investment can spread capability beyond the organisations able to own such systems. Efficient shared compute then becomes both a research asset and a tool for broader economic innovation.
Industrial access also creates a varied workload. One company may need simulation, another model training and a third large-scale data analysis. The centre must help users choose methods that fit the machine instead of assuming every problem benefits equally from GPU acceleration. Good project selection protects scarce capacity and makes outcomes easier to demonstrate.
The warm-water system offers an engineering lesson beyond Daresbury. As compute density rises, cooling must be designed as part of the machine rather than as an afterthought surrounding it. Reducing energy demand by up to 40% can influence facility cost and environmental impact, while the reported performance improvement shows that thermal management can support speed as well as efficiency.
Public investment can also create evidence for future infrastructure decisions. Usage patterns, energy consumption and project outcomes will show where demand is strongest and which forms of support businesses need. The Hartree Centre can use that evidence to refine training and access, ensuring that the new machine functions as a national capability rather than an impressive but underused installation.
