The UK Atomic Energy Authority (UKAEA) is installing a new supercomputer named Sunrise, which is specifically designed to support nuclear fusion research. Based on AMD Epyc and Instinct-powered Dell hardware, Sunrise achieves approximately 6 exaflops at
at 8-bit precision. That means it is capable of running six quintillion floating point operations per second.
While it is not as powerful as Bristol’s Isambard-AI supercomputer, which went live in June 2025, Sunrise is being positioned as the most powerful computer of its kind specifically designed for nuclear fusion research.
“Sunrise effectively allows us to take on a moonshot-like problem a lot more cost effectively, reduce risk and accelerate the time of delivery of commercial fusion devices,” Rob Akers, UKAEA director for computing programmes, told Computer Weekly.
Built on AMD MI 355X graphics processor units (GPUs), Sunrise is being optimised for the intersection of AI and high-performance computing (HPC).
At 64-bit precision, Akers says Sunrise delivers around 50 petaflops (50 trillion floating point operations per second), which means it can run high-fidelity simulations. According to Akers, the combined performance of 8-bit and 64-bit precision available on Sunrise makes it highly suitable for AI applications and inference tasks. He says: “It’s an incredibly powerful machine, tuned to do some very interesting work.”
Eight bits used to be state of the art in the heyday of home computing in the late 1970s and early 1980s. But it is now a figure that is very relevant in today’s high performance computing. When asked why the 8-bit precision figure matters, given its 64-bit precision figure of 50 petaflops, Akers says: “High performance computing floating point precision is not the best metric for an AI machine.”
Instead, an AI machine’s power is a measure of its ability to process tokens such as read individual characters in a text file to extract meaning as Akers explains: “The interesting thing is that 8-bit precision has become an incredibly powerful part of the computing landscape now because of large language models (LLMs). We are going to be doing work in that space, building very bespoke models that will ingest text such as document archives that have been collected over many, many decades and turn that into useful information and knowledge.”
He says Sunrise will use AI to help solve complex engineering problems in a cost-effective way. “Without AI, the problems we are trying to solve would be intractable,” he says. It will be used to create surrogate models that can work on a workstation and be able to simplify high-fidelity simulations, allowing them to run efficiently on smaller systems.
For instance, digital twins running on Sunrise will be used to simulate multi-scale physics and coupled systems, which Akers says, means nuclear fusion researchers will able to address emergent behaviours and “black swan” events in fusion power plants. “Digital twins give us the ability to simulate very high-fidelity models, and very importantly, strongly coupled models,” he says. “In a fusion power plant, there are lots of different physical mechanisms that couple the plant together—everything from structural forces due to gravity, but also due to electromagnetism and then there’s a lot of heat flow and radiation flow across the system. Everything’s coupled together.”
By using Sunrise to run digital twin simulations, the UKAEA aims to improve confidence in designs and enable breakthroughs in machinery development for nuclear fusion.
