Exascale computing capability by 2030 and significant investment in compute power designed to support AI research are among the most high-profile recommendations in a new review of UK compute power needs published by the government today. The Future of Compute Report was launched alongside details of a new £370m Science and Technology Framework. Developing sovereign capability could aid UK researchers who currently have to rely on the resources of the US hyperscale cloud providers.
Supercomputers passed the exascale mark in June 2020 when Japan’s Fugaku achieved 1.42 exaFLOPS. The first true exascale computer, Frontier, was launched in June 2022. There are multiple exascale supercomputers under development around the world.
At the same time, one of the biggest bottlenecks in artificial intelligence research, particularly around the development of complex large language models (LLM) is compute power. The new report comes off the back of calls from some industry and academic leaders for a sovereign LLM for the UK.
Prime Minister Rishi Sunak wants to help the UK cement its place as a global science and technology superpower by 2030, including through quantum computing and AI but to achieve those goals the country needs to up its compute power game, or at least improve the way it is organised, that is according to the new paper.
It includes analysis and research on current and future requirements for compute power to achieve Sunak’s goal, and is written by experts including Google VP of research and University of Cambridge Professor Zoubin Ghahramani, who led the group.
It found that as it stands the UK “lacks a long-term vision for compute”, suffering from a complex ecosystem that is hard for users to navigate. “Existing compute capabilities are not fit to serve all users, particularly AI users, and are falling behind those of other advanced economies,” it says.
Compute power for AI training and research
There were several use cases listed including cutting-edge computational research, large-scale modelling, simulations and data science, small-scale modelling and AI training and research listed in the report as requirements for better compute power infrastructure. “The UK’s compute ecosystem needs to reflect the variety of users, both existing and emerging, and their needs,” the authors declared. “It needs to ensure a broad offering to support prosperity and growth across the UK.”
The problem is the current provision doesn’t meet these needs which the authors say is “limiting the UK’s scientific capability and inhibiting scientific breakthroughs”. Beyond academia, the authors say it could also benefit SMEs and lead to economic growth.
The key recommendations in this area include delivering full exascale capability by 2026, establishing a lighthouse project to develop interoperability solutions, establish a UK AI research resource by this summer including 3,000 top-spec AI accelerators by 2023 and developing comprehensive planning guidelines for sustainable compute infrastructure.
“The UK’s ambitions are for long-term economic growth and to cement its status as a science and technology superpower by 2030,” the authors wrote. “These ambitions rely on the UK’s ability to harness the opportunities that compute provides and maximise the impact of this technology across its economy and science base. The government must be ambitious and visionary to truly use compute to deliver its objectives.”
Ekaterina Almasque, general partner at deep technology venture capital company OpenOcean, said the plea to support the compute ecosystem through public investment will have support across the technology sector, explaining that creating such an ecosystem with access to high-performance computing will make it easier for start-ups and founders to bring resources to bear in exploring new frontiers like quantum computing, as well as develop AI solutions and tools.
“Compute is the engine room of fields like AI and quantum,” she said. “By investing in the infrastructure of the data economy — the essential building blocks — we can provide the technology needed to support data-intensive processes like drug discovery, chemistry, and large-scale optimisations needed in finance and logistics. This enables our domestic companies to scale, compete with international rivals, and support the UK’s ambitions as a tech superpower.”
Providing support for start-ups
To make it work the government must also ensure that start-ups can access enough training data to feed exascale computers, which will allow them to compete with hyperscalers like Google and Microsoft, which have a significant stable of proprietary data.
Mark Boost, CEO of cloud-native service provider Civo described the review as a “shot in the arm for the UK compute space” that businesses have been crying out for. “I particularly welcome the review’s focus on making the cloud an affordable route to access compute resources,” he says. “For too long, hyperscalers have over-promised and under-delivered for users, offering an unreliable service with ballooning prices that raises the bar for access.”
Creating a national sovereign high-performance computing infrastructure would also help develop alternatives to the cloud hyperscalers, says Alan Turing Institute Director of Foundational AI Research and University of Oxford professor Michael Wooldridge. He believes the dominance of Amazon’s AWS, Microsoft Azure and Google Cloud is the biggest problem in compute today.
“To gain access to them, researchers in the public sector have to sign non-disclosure agreements and go behind closed doors,” Wooldridge says. “This is a deeply worrying state of affairs: the most important technology of the 21st century should not be the exclusive preserve of big tech.”
Having better access to compute power could then lead to advancements in AI as many tools require specialised supercomputers to build and run. “The report recognises all of this and sets out a roadmap to take us forward,” Wooldridge adds. “This is fantastic, but we need to be aware this is going to be an ongoing challenge – it isn’t something we can fix and then forget about. It’s going to be with us for the foreseeable future.”
He said it would also help the UK develop its own sovereign AI capability, explaining that the Alan Turing Institute believes it would be “utterly unacceptable for government organisations who want to use AI to feed sensitive UK data to the algorithms of Big Tech companies. The single biggest hurdle to overcome in developing a sovereign AI capability is our national lack of compute resources – and the report therefore neatly dovetails with our proposals.”
Read more: This is how GPT-4 will be regulated
