UK Bets On Intel CPUs And GPUs, Dell Boxen, OpenStack For Dawn Supercomputer

Intel, Dell, and the University of Cambridge are today talking up Dawn, a UK supercomputer that's being deployed as you read this.

Dawn, when complete and running, will be part of the British government's so-called AI Research Resource (AIRR), and was developed with investment from the govt's Research and Innovation (UKRI) unit. The machine will be put to work on demanding workloads including academic and industrial research, healthcare, engineering, and climate modelling.

Somewhat tediously, Intel said Dawn will be the "UK's fastest AI supercomputer" while Nvidia, which is behind Isambard-AI, another upcoming UK super, says its machine will be the country's "most powerful." We know Isambard-AI is aiming for 200 petaFLOPS of (presumably theoretical peak) FP64 performance, and we don't know Dawn's figures. The full technical and performance details of the latter computer are under wraps until the SC23 supercomputing conference in the US later this month. We'll have to wait to see which one is the real winner.

The UK's current publicly known champion machine Archer2 is 30th in the latest global top-500 rankings and theoretically peaks at 26PFLOPS (20PFLOPS in benchmarks) with FP64. Isambard-AI is set to be about ten times faster than Archer2.

Dawn’s project team said their machine, located at the Cambridge Open ZettaScale Lab, will take the UK closer to reaching its goal of having its own exascale system. The team has also stated Dawn will be among the most powerful publicly known systems in Europe.

How Dawn will compare to the LUMI supercomputer in Finland, which is currently Europe's fastest clocking in at 429 petaFLOPS peak (309PFLOPS in benchmarks) with FP64, remains to be seen.

Dawn is also due to be rolled out in two phases: the first over the next two months, and the second phase next year. It’s unclear to us if Intel's “fastest” claim pertains to the first phase or to the second phase, the latter of which is supposed to deliver ten times the performance of Dawn’s first rising.

"Dawn Phase 1 represents a huge step forward in AI and simulation capability for the UK, deployed and ready to use now," said Dr Paul Calleja, director of Research Computing Services at the University of Cambridge, in a canned statement.

"If taken forward, Dawn Phase 2 would significantly boost the UK's AI capability and continue this successful industry partnership," he added, which would seem to suggest that Phase 2 is not yet a done deal.

Dawn is made up of Dell’s PowerEdge XE9640 servers, 2U boxen fitted with a pair of Intel 4th Gen Xeon Scalable processors and four of Intel’s Data Center GPU Max accelerators. Liquid cooling will keep those rigs from bubbling over with computational excitement.

But it seems we shall have to wait for SC23 to learn of details like server counts and memory capability, and how the boxes are interconnected.

Dawn will run Scientific OpenStack, a variant of the open source cloud platform that's supported by Brit developer StackHPC. Intel's oneAPI programming model that supports heterogeneous computing will also be in the mix.

• Tachyum says someone will build 50 exaFLOPS super with its as-yet unfinished chips
• US govt talks up $2B X-ray photobooth to check its nuke weapon sims are right
• Atos subsidiary Eviden scores contract win in Europe's first exascale system
• From vacuum tubes to qubits – is quantum computing destined to repeat history?

Dawn Phase 1 and the Isambard-AI supercomputer operated by the University of Bristol will join to form the aforementioned AIRR, a UK national facility to help researchers work on AI projects.

According to HPE, which is building Isambard-AI, this super will be based on the Cray EX supercomputer architecture and contain 5,448 Nvidia GH200 Grace Hopper Superchips, which combine the Arm-based Grace CPU with a Hopper-based GPU. Construction on that £225 million ($273 million) system is due to start next year.

"Collaborations like this one between the University of Cambridge, Dell Technologies and Intel, alongside strong inward investment, are vital if we want compute to unlock the high-growth AI potential of the UK," said Dell head of UK Public Sector Tariq Hussain.

"It is paramount that the government invests in the right technologies and infrastructure to ensure the UK leads in AI and exascale-class simulation capability," he added.

When plans for the UK's first exascale computer were whispered earlier this year, it was suggested the machine would be an all-British affair, built by UK firms and using chips and systems provided by the nation's businesses, such as AI startup Graphcore. But with Graphcore facing mounting losses and needing new investment to finance existing commitments, this now seems unlikely. ®