The first exascale supercomputer will be hosted by the University of Edinburgh, the UK government has confirmed. This next-generation computing system will be able to handle at least a quintillion (or, a billion billion) calculations per second and be 50 times faster than the current generation of supercomputers currently running in the UK.
Exascale computing has become the latest battleground in the race for global compute dominance, with the US becoming the first country to successfully build an exascale supercomputer in 2022. Named “Frontier” and housed at the Oak Ridge National Laboratory, the machine is comprised of 9,400 AMD-powered nodes and 90 miles of networking cables, allowing it to run at 1.1 exaflops.
UK Prime Minister Rishi Sunak announced plans to invest in high-performance computing power ahead of the government’s Spring budget earlier this year. This funding boost has already backed the construction of a £900m AI supercomputer in Bristol in addition to the new exascale general purpose machine bound for Edinburgh.
The University of Edinburgh already hosts a supercomputing centre and this will be expanded as part of the new £31m national exascale facility. The new supercomputing system will be funded by the Department for Innovation, Science and Technology (DSIT) and include chips and components built by British companies. The manufacturers and designers have not been revealed yet, but Sunak said that future computing projects funded by the government would include domestically manufactured chips.
Work will begin in 2025
Exascale machines are built to carry out complex functions at faster speeds and with more precision than the current generation of supercomputers. In time, it is hoped that the construction of these much larger and more powerful machines will allow researchers and companies to facilitate new breakthroughs in the fields of AI, medicine, nuclear fusion, chemistry, and climate change.
The UK’s current national supercomputer, ARCHER2, is run at the University of Edinburgh by EPCC (formerly the Edinburgh Parallel Computing Centre) and came online in November 2021. The machine has a peak performance of 28 petaflops, or about 28 billion calculations per second, and boasts 750,000 cores, including 7,742 AMD EPYC 64-core processors. The HPE Cray EX system cost £79m and is 10 billion times faster than the first Cray supercomputer built in 1964.
The new, as yet unnamed exascale machine will also be operated by EPCC. The first installation phase is expected to begin in 2025, though it remains unclear as to when the machine will be fully operational. “These supercomputers are immensely complex systems, and we’ll use everything we’ve learnt over the past 30 years to run the best possible service for our thousands of users from across the UK’s scientific and industrial research communities,” said the EPCC’s director, Professor Mark Parsons.
Science, Innovation and Technology Secretary Michelle Donelan said exascale machines are required to keep the UK as a global leader in scientific discovery and innovation. Donelan argued that the investment “will provide British researchers with an ultra-fast, versatile resource to support pioneering work into AI safety, life-saving drugs, and clean low-carbon energy.”
UK Research and Innovation Chief Executive Professor Dame Ottoline Leyser added that state-of-the-art compute infrastructure is critical to unlock advances in research and innovation including drug design and energy security. “This next phase of investment,” said Leyser, “will help to keep the UK at the forefront of emerging technologies and facilitate the collaborations needed to explore and develop game-changing insights across disciplines.”
As well as plans for three exascale machines across the US, the European Union is also heavily investing in this new generation of supercomputing machine. EuroHPC is leading the EU project and the first machine will be built at the Julich Supercomputing Centre in Germany for an estimated €500m. Named JUPITER, it is expected to be up and running by mid-2024 with more than two exaflops of processing power available.