Cardiff University and ARCCA
Advanced Research Computing at Cardiff
The team that runs Cardiff University's supercomputing services, including Falcon.
ARCCA
Who we are
Advanced Research Computing at Cardiff (ARCCA) is Cardiff University’s high performance computing team. It runs the University’s supercomputing services, including Falcon (the successor to Hawk), and gives researchers expert advice and training. ARCCA is led by Director Dr Jon Lockley and governed by the Cardiff Operational and User Group, which focuses on user needs and HPC development.
ARCCA supports research across the University. Its traditional strengths are computational chemistry, materials and engineering. Falcon opens the same platform to bioscience, AI and large-scale simulation, and some researchers have funded expansions to the service themselves.
Beyond Cardiff, ARCCA plays a key role in Supercomputing Wales and runs its Cardiff hub. It is also a partner in Isambard 3, the GW4 Tier 2 service, and in the Durham-led SHAREing project.
Research impact
What supercomputing at Cardiff delivers
The system
Falcon
In 2023, Cardiff committed £5.5m to refresh its HPC data centre and update Hawk, so the service could keep pace with research in complex data and AI. Falcon is the result: a CPU-focused system with GPUs and high performance networking for simulation-led AI. It combines new CPU and GPU nodes with Hawk’s AMD Rome compute nodes and NVIDIA V100 GPU nodes, and it has grown since it was installed. Falcon is housed at Cloud Centres’ Tier III-certified data centre north of Cardiff, in a suite built for direct liquid cooling.
As of November 2026, Falcon has 711 registered users, 400 of them active. More than 30% of active projects use AI tools.
See The Solution page for Falcon’s full configuration.
Sim2Real
From simulation to real-world AI
Research at Cardiff is moving towards “Sim2Real”: high-precision (FP64) simulation that produces the data needed to train AI models that work reliably in the real world. This needs the numerical precision of scientific simulation and the tensor processing power of deep learning together, and standard AI clusters don’t offer both. Today, researchers can’t run high-fidelity simulations fast enough to produce that training data. Sim2Real is the only viable way past this bottleneck in priority areas including quantum, net zero and economic growth.
Initial workloads will use native FP64 simulation, alongside an evaluation of FP64 emulation on low-precision tensor cores, which could improve throughput and energy efficiency. Porting existing applications will show the trade-off between performance and accuracy, and where it makes sense, emulated components will be replaced with alternative methods. Manufacturers are moving towards lower-precision silicon for AI, so every researcher using accelerators needs to know whether emulation affects scientific validity.
The same equipment lets Cardiff run multidisciplinary work across biosciences, engineering, chemistry, environmental studies and computational sciences. That reduces inefficiencies, strengthens collaboration and supports regional centres of excellence built on Cardiff’s AI leadership and training programmes. Training AI models locally before scaling up to national facilities such as Dawn and Isambard-AI shows they are ready for time on those systems. ARCCA is also moving other applications, such as genomics, onto GPUs.