The University of Bristol’s Islamabad-AI supercomputer is now online. Supported by £225m of government funding, the machine will be used by organisations such as the UK’s AI Safety Institute for AI research from this month.
The funding, which is part of a £300m package to create a national Artificial Intelligence Research Resource (AIRR) for the country, was announced at the government’s AI Safety Summit at Bletchley Park in October 2023.
The Isambard-AI phase one is based on a Cray EX2500 supercomputer with 168 Nvidia GH200 GraceHopper “superchips”.
Islambad-AI will connect with another new supercomputer cluster in Cambridge, called Dawn, which is being co-designed by Intel, Dell Technologies and the University of Cambridge.
The hardware at Bristol includes Hewlett Packard Enterprise (HPE) Slingshot 11 interconnect and direct liquid cooling with highly integrated, heterogeneous CPU-GPU systems from Nvidia. According to HP, it is one of the most efficient supercomputers that has ever been built.
Isambard-AI phase one’s performance is around 7.4 PFLOP/s, which puts it at position 129 in the latest edition of the Top500 list of high-performance computing (HPC) systems on Earth, as published at the ISC High Performance 2024 event in Hamburg, Germany.
Simon McIntosh-Smith, director of the Bristol Centre for Supercomputing at the University of Bristol, said: “Isambard-AI phase 1 signifies the start of the Isambard-AI service. When the remaining 5,280 GPUs arrive at the University’s National Composites Centre (NCC) later in the summer, it will increase the performance by a factor of 32.”
Minister for AI, Jonathan Berry, said: “With the launch of the first stage of University of Bristol’s supercomputer Isambard-AI, we’re witnessing a groundbreaking moment for UK science, innovation and technology. This world-class equipment will revolutionise research possibilities here in the UK, with our world-first AI Safety Institute among the organisations who are set to benefit by harnessing one of the most powerful computer systems on the planet to drive forward their vital research.
“Not only does Isambard-AI rank among the world’s fastest supercomputers, but it also sets the standard for eco-conscious computing, leading the charge in efficiency and sustainability. From AI safety to healthcare and climate research, its capabilities are unparalleled, marking the UK out as a global leader in AI and responsible innovation.”
While the performance of Islambad-AI is a major step forward in UK supercomputing, at ISC High Performance 2024, Intel announced in collaboration with Argonne National Laboratory and HPE that the Aurora supercomputer has broken the exascale barrier at 1.012 exaflops and is the fastest AI system in the world dedicated to AI for open science, achieving 10.6 AI exaflops.
Islambad-AI is also second on the Green500, behind Jedi, the first module of the exascale supercomputer Jupiter, which is based on Eviden’s latest BullSequana XH3000 architecture.
While Islamabad-AI and Jedi use Grace Hopper superchips, Jedi’s watercooling uses a hot water-cooling system called Direct Liquid Cooling, which claims to require significantly less energy than conventional air cooling, and allows the heat generated to be reused downstream.