̽��ֱ�� of Cambridge - high-performance computing

Cambridge, Intel and Dell join forces on UK’s fastest AI supercomputer

cjb250 — Wed, 01 Nov 2023 10:51:14 +0000

Dawn has been created via a highly innovative long-term co-design partnership between the ̽��ֱ�� of Cambridge, UK Research & Innovation, the UK Atomic Energy Authority and global tech leaders Intel and Dell Technologies. This partnership brings highly valuable technology first-mover status and inward investment into the UK technology sector.

Dawn, supported by UK Research and Innovation (UKRI), will vastly increase the country's AI and simulation compute capacity for both fundamental research and industrial use, accelerating research discovery and driving growth within the UK knowledge economy. It is expected to drive significant advancements in healthcare, green fusion energy development and climate modelling.

Dawn Phase 1 and the already announced Isambard AI supercomputer at the ̽��ֱ�� of Bristol will join to form the AI Research Resource (AIRR), a UK national facility to help researchers maximise the potential of AI and support critical work into the potential and safe use of the technology.

Dr Paul Calleja, Director of Research Computing Services at the ̽��ֱ�� of Cambridge, said: “Dawn Phase 1 represents a huge step forward in AI and simulation capability for the UK, deployed and ready to use now. Dawn was born from an innovative co-design partnership between ̽��ֱ�� of Cambridge, UKAEA, Dell Technologies and Intel.

“ ̽��ֱ��Phase 1 system plays an important role within a larger context, where this co-design activity is hoped to continue, aiming to deliver a Phase 2 supercomputer in 2024 which will boast 10 times the level of performance. If taken forward, Dawn Phase 2 would significantly boost the UK AI capability and continue this successful industry partnership.”

World-leading technical teams from the ̽��ֱ��, Intel and Dell Technologies built Dawn, which harnesses the power of both AI and high-performance computing (HPC) to solve some of the world’s most challenging and pressing problems.

Announcing this investment at the AI Safety Summit at Bletchley Park, Science, Innovation and Technology Secretary Michelle Donelan said: "Frontier AI models are becoming exponentially more powerful. At our AI Safety Summit in Bletchley Park, we have made it clear that Britain is grasping the opportunity to lead the world in adopting this technology safely so we can put it to work and lead healthier, easier and longer lives.

"This means giving Britain’s leading researchers and scientific talent access to the tools they need to delve into how this complicated technology works. That is why we are investing in building UK’s supercomputers, making sure we cement our place as a world-leader in AI safety."

Professor Emily Shuckburgh, Director of Cambridge Zero and the Institute of Computing for Climate Science said: “ ̽��ֱ��coupling of AI and simulation methods is a growing and increasingly essential part of climate research. It is central to data-driven predictions and equation discovery, both of which are at the fore in climate science.

“This incredible new resource – Dawn – at Cambridge will enable software engineers and researchers at the Institute of Computing for Climate Science to accelerate their work helping to address the global challenges associated with climate change.”

Dawn brings the UK closer to reaching the compute threshold of a quintillion floating point operations per second – one exaflop, better known as exascale. For perspective: every person on earth would have to make calculations 24 hours a day for more than four years to equal a second’s worth of processing power in an exascale system.

Hosted at Cambridge Open Zettascale Lab’s site, Dawn is the fastest AI supercomputer deployed in the UK today and will support some of the UK’s largest-ever workloads across both academic research and industrial domains. Importantly, it is the UK's first step on the road to developing future Exascale system.

Adam Roe, EMEA HPC technical director at Intel, said: “Dawn considerably strengthens the scientific and AI compute capability available in the UK, and it’s on the ground, operational today at the Cambridge Open Zettascale Lab.

“I’m very excited to see the sorts of early science this machine can deliver and continue to strengthen the Open Zettascale Lab partnership between Dell Technologies, Intel and the ̽��ֱ�� of Cambridge, and further broaden that to the UK scientific and AI community.”

Tariq Hussain, Head of UK Public Sales, Dell Technologies, said: "Collaborations like [this one], alongside strong inward investment, are vital if we want compute to unlock the high-growth AI potential of the UK. 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.

“It's also important to embrace the full spectrum of the technology ecosystem, including GPU diversity, to ensure customers can tackle the growing demands of generative AI, industrial simulation modelling and ground-breaking scientific research."

Dr Rob Akers, Director of Computing Programmes & Senior Fellow at UKAEA, added: “Dawn will form an essential part of a diverse UKRI supercomputing ecosystem, helping to promote high-fidelity simulation and AI capability ensuring that UK science and engineering is first in the queue to exploit the latest innovation in disruptive HPC hardware. In the short term, Dawn will allow UKAEA’s fusion energy programme to form a powerful and exciting cross-Atlantic partnership with US labs exploiting the new 2ExaFlop AURORA supercomputer at Argonne, Dawn's ‘big sister’.

“Fusion has long been referred to as an ‘exascale grand challenge’. ̽��ֱ��exascale is finally upon us and I firmly believe that the many collaborations coalescing around Dawn will be a powerful ingredient for extracting value promised by the exascale – for the UK to deliver fusion power to grid in the 2040s, to realise Net Zero more generally, to seed high value UK jobs in AI and ‘digital’ and to drive economic growth across the entire United Kingdom.”

̽��ֱ��Cambridge Open Zettascale Lab is hosting Dawn, the UK’s fastest artificial intelligence (AI) supercomputer, which has been built by the ̽��ֱ�� of Cambridge Research Computing Services, Intel and Dell Technologies.

Dawn Phase 1 represents a huge step forward in AI and simulation capability for the UK, deployed and ready to use now

Paul Calleja

Joe Bishop

Dr Paul Calleja, Director of Dawn AI Service (left) and Professor Richard McMahon, Chair of Cambridge Research Computing Advisory Group and UKRI Dawn Principal Investigator (right) in front of Dawn.

̽��ֱ��text in this work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. Images, including our videos, are Copyright © ̽��ֱ�� of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways – as here, on our main website under its Terms and conditions, and on a range of channels including social media that permit your use and sharing of our content under their respective Terms.

Yes

Licence type:

Attribution

Design work on ‘brain’ of world’s largest radio telescope completed

sc604 — Thu, 09 May 2019 09:47:47 +0000

̽��ֱ��SKA’s Science Data Processor (SDP) consortium has concluded its engineering design work, marking the end of five years’ work to design one of two supercomputers that will process the enormous amounts of data produced by the SKA’s telescopes.

̽��ֱ��SDP consortium, led by the ̽��ֱ�� of Cambridge, has designed the elements that will together form the ‘brain’ of the SKA. SDP is the second stage of processing for the masses of digitised astronomical signals collected by the telescope’s receivers. In total, close to 40 institutions in 11 countries took part.

̽��ֱ��UK government, through the Science and Technology Facilities Council (STFC), has committed £100m to the construction of the SKA and the SKA Headquarters, as its share as a core member of the project. ̽��ֱ��global headquarters of the SKA Organisation are located in the UK at Jodrell Bank, home to the iconic Lovell Telescope

“It’s been a real pleasure to work with such an international team of experts, from radio astronomy but also the High-Performance Computing industry,” said Maurizio Miccolis, SDP’s Project Manager for the SKA Organisation. “We’ve worked with almost every SKA country to make this happen, which goes to show how hard what we’re trying to do is.”

̽��ֱ��role of the consortium was to design the computing hardware platforms, software, and algorithms needed to process science data from the Central Signal Processor (CSP) into science data products.

“SDP is where data becomes information,” said Rosie Bolton, Data Centre Scientist for the SKA Organisation. “This is where we start making sense of the data and produce detailed astronomical images of the sky.”

To do this, SDP will need to ingest the data and move it through data reduction pipelines at staggering speeds, to then form data packages that will be copied and distributed to a global network of regional centres where it will be accessed by scientists around the world.

SDP itself will be composed of two supercomputers, one located in Cape Town, South Africa and one in Perth, Australia.

“We estimate SDP’s total compute power to be around 250 PFlops – that’s 25% faster than IBM’s Summit, the current fastest supercomputer in the world,” said Maurizio. “In total, up to 600 petabytes of data will be distributed around the world every year from SDP –enough to fill more than a million average laptops.”

Additionally, because of the sheer quantity of data flowing into SDP: some 5 Tb/s, or 100,000 times faster than the projected global average broadband speed in 2022, it will need to make decisions on its own in almost real-time about what is noise and what is worthwhile data to keep.

̽��ֱ��team also designed SDP so that it can detect and remove manmade radio frequency interference (RFI) – for example from satellites and other sources – from the data.

“By pushing what’s technologically feasible and developing new software and architecture for our HPC needs, we also create opportunities to develop applications in other fields,” said Maurizio.

High-Performance Computing plays an increasingly vital role in enabling research in fields such as weather forecasting, climate research, drug development and many others where cutting-edge modelling and simulations are essential.

Professor Paul Alexander, Consortium Lead from Cambridge’s Cavendish Laboratory said: “I’d like to thank everyone involved in the consortium for their hard work over the years. Designing this supercomputer wouldn’t have been possible without such an international collaboration behind it.”

An international group of scientists led by the ̽��ֱ�� of Cambridge has finished designing the ‘brain’ of the Square Kilometre Array (SKA), the world’s largest radio telescope. When complete, the SKA will enable astronomers to monitor the sky in unprecedented detail and survey the entire sky much faster than any system currently in existence.

Designing this supercomputer wouldn’t have been possible without such an international collaboration behind it

Paul Alexander

SKA Organisation

Artist’s impression of the full Square Kilometre Array at night

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. Images, including our videos, are Copyright © ̽��ֱ�� of Cambridge and licensors/contributors as identified. All rights reserved. We make our image and video content available in a number of ways – as here, on our main website under its Terms and conditions, and on a range of channels including social media that permit your use and sharing of our content under their respective Terms.

Yes

̽��ֱ��Electron Manifesto: transforming high performance computing with 'spintronics'

tdk25 — Wed, 26 Jul 2017 11:54:19 +0000

In the early days of the computer, calculators were room-sized and public demand was low. Now, it’s the reverse. Digital technology has become smaller and faster, and our dependence on it has grown.

We are almost desensitised to a stream of facts about the startling rate at which this is occurring. In 2016, IBM found that humans now create 2.5 quintillion bytes of data daily. From the start of this decade to its end, the world’s data will increase 50 times over.

̽��ֱ��basic building blocks of electronic devices, such as the transistor, work by moving packets of charge around a circuit. A single unit of charge is an electron, and its movement is governed by semiconductors, commonly made from silicon. But technology based on these principles is now reaching a point where it cannot get much smaller or faster. A paradigm shift is due.

“There have been many failed attempts to oust silicon from its predominance,” reflects Professor Mark Blamire, Head of Materials Science at Cambridge. “Something has to be done because the technology can’t be scaled to smaller sizes for very much longer. It’s already a major source of power consumption. There’s no obvious competitor, so in a sense the opportunity is there.”

Blamire and his colleague Dr Jason Robinson are leading several major programmes investigating one such competitor, known as superconducting spintronics.

̽��ֱ��launch of a UK-based programme last year provoked excitement within the scientific community. “Cambridge Uni spins up green and beefy supercomputer project,” announced British tech site ̽��ֱ��Register, for example. One reason in particular is because superconducting spintronics might address the eye-watering energy consumption of the huge server farms that handle internet traffic. Data centres account for 3% of the world’s electricity supply and about 2% of greenhouse gas emissions.

̽��ֱ��project combines two phenomena: superconductivity and spin. Superconductivity refers to the fact that at low temperatures some materials carry a charge with zero resistance. Unlike, for example, copper wires, which lose energy as heat, superconductors are therefore extremely energy efficient.

‘Spin’ is the expression for electrons’ intrinsic source of magnetism. Originally it was thought that this existed because electrons were indeed spinning, which turned out to be wrong, but the name stuck, and it is still used to describe the property in particles that makes them behave a bit like tiny bar magnets. Like a magnet, this property makes the electrons point a certain way; the spin state is therefore referred to as ‘up’ or ‘down’.

Researchers have been using the magnetic moments of electrons to store and read data since the 1980s. At their most basic, spintronic devices use the up/down states instead of the 0 and 1 in conventional computer logic.

Spintronics could also transform the way in which computers process information. ̽��ֱ��researchers envisage that instead of the devices moving packets of charge around, they will transmit information using the relative spin of a series of electrons, known as a ‘pure spin current’, and sense these using magnetic elements within a circuit.

By eliminating the movement of charge, any such device would need less power and be less prone to overheating – removing some of the most significant obstacles to further improving computer efficiency. Spintronics could therefore give us faster, energy-efficient computers, capable of performing more complex operations than at present.

To generate large enough spin currents for memory and logic devices, significant charge is required as an input, and the power requirements of this currently outweigh many of the benefits. Using a superconductor to provide that charge, given its energy efficiency, would present a solution. But the magnetic materials used to control spin within spintronic devices also interfere with superconductivity.

This problem was thought insurmountable until, in 2010, Robinson discovered how to combine superconductors and spintronics so that they can work together in complete synergy. His team added an intervening magnetic layer (a material called holmium). By using this interface, they were able to preserve the delicate balance of electron pairing that’s needed to achieve superconductivity, but still managed to create a bias within the overall spin of the electrons.

This, explains Robinson, “created a marriage that opens up the emerging field of superconducting spintronics.” Over the next five years, he and Blamire developed the field, and last year were awarded a major grant from the Engineering and Physical Sciences Research Council: “To lead the world in understanding the coupling of magnetism and superconductivity to enable future low energy computing technologies.” Robinson has since been awarded a second grant with Professor Yoshi Maeno, from the ̽��ֱ�� of Kyoto, to broaden materials research on superconducting spintronics.

Although still at an experimental stage, the project – which includes collaborators from Imperial College London, ̽��ֱ�� College London and Royal Holloway London – is tackling questions such as how to generate and control the flow of spin in a superconducting system. And its scope is already expanding. “We have found more ways of achieving what we are trying to do than we originally dreamed up,” Robinson says.

One example involves making potentially innovative use of superconductivity itself. In ‘conventional’ spintronics, spin is manipulated through the interactions between magnetic materials within the device. But Blamire has found that when a superconductor is placed between two ferromagnets, its intrinsic energy depends on the orientation of those magnetic layers. “Turning that on its head, if you can manipulate the superconducting state, you can control the orientation of the magnetic layers, and therefore the spin,” he says.

Meanwhile, Robinson has led a study that for the first time enabled graphene, a material already recognised for its potential to revolutionise the electronics industry, to superconduct. This raises the possibility of using this extraordinary material, and other two-dimensional materials like it, in superconducting spintronics.

Although approaches like this are still being tested, Blamire says that by 2021 the team will have developed sample logic and memory devices that fuse superconductivity and spin. These proof-of-concept models could, perhaps, be incorporated into a new type of computer processor. “It would be a huge step to get from there to a device that could be competitive,” he admits. “It’s not necessarily difficult, but it would require considerable investment.”

̽��ֱ��project is set up to enable industrial collaboration in the years to come. A key partner is the Hitachi Lab in Cambridge, while the project’s advisory board also features representatives from the Cambridge-based semiconductor firm ARM, and HYPRES, a digital superconductor company in the USA.

Robinson points out that the UK – and Cambridge in particular – has historical strengths in research into superconductivity and spintronics, but adds that a “grand challenge” has long been needed to focus academic investigation on a meaningful partnership with industry.

Leading low-energy computing into a post-semiconductor age is certainly grand. Silicon’s domination, after all, stretches from its eponymous valley in California, to a fen in Cambridge, a gulf in the Philippines and an island in Japan.

Can the unlikely – not to say still primitive – marriage of spintronics and superconductivity really replace an electronic empire on which the sun never sets? “I suspect people had similar questions at the dawn of the semiconductor,” Robinson observes. “One shouldn’t lose sight of what we are doing here. We aren’t just trying to do something better; we are offering something entirely different and new.”

Electron ‘spin’ could hold the key to managing the world’s growing data demands without consuming huge amounts of energy. Now, researchers have shown that energy-efficient superconductors can power devices designed to achieve this. What once seemed an impossible marriage of superconductivity and spin may be about to transform high performance computing.

One shouldn’t lose sight of what we are doing here. We aren’t just trying to do something better; we are offering something entirely different and new.

Jason Robinson

Creativity103

Spinning top

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. For image use please see separate credits above.

Yes

Licence type:

Attribution-ShareAlike

Cambridge to research future computing tech that could “ignite a technology field”

tdk25 — Thu, 14 Apr 2016 23:01:34 +0000

A project which aims to establish the UK as an international leader in the development of “superconducting spintronics” – technology that could significantly increase the energy-efficiency of data centres and high-performance computing – has been announced.

Led by researchers at the ̽��ֱ�� of Cambridge, the “Superspin” project aims to develop prototype devices that will pave the way for a new generation of ultra-low power supercomputers, capable of processing vast amounts of data, but at a fraction of the huge energy consumption of comparable facilities at the moment.

As more economic and cultural activity moves online, the data centres which house the servers needed to handle internet traffic are consuming increasing amounts of energy. An estimated three per cent of power generated in Europe is, for example, already used by data centres, which act as repositories for billions of gigabytes of information.

Superconducting spintronics is a new field of scientific investigation that has only emerged in the last few years. Researchers now believe that it could offer a pathway to solving the energy demands posed by high performance computing.

As the name suggests, it combines superconducting materials – which can carry a current without losing energy as heat – with spintronic devices. These are devices which manipulate a feature of electrons known as their “spin”, and are capable of processing large amounts of information very quickly.

Given the energy-efficiency of superconductors, combining the two sounds like a natural marriage, but until recently it was also thought to be completely impossible. Most spintronic devices have magnetic elements, and this magnetism prevents superconductivity, and hence reduces any energy-efficiency benefits.

Stemming from the discovery of spin polarized supercurrents in 2010 at the ̽��ֱ�� of Cambridge, recent research, along with that of other institutions, has however shown that it is possible to power spintronic devices with a superconductor. ̽��ֱ��aim of the new £2.7 million project, which is being funded by the Engineering and Physical Sciences Research Council, is to use this as the basis for a new style of computing architecture.

Although work is already underway in several other countries to exploit superconducting spintronics, the Superspin project is unprecedented in terms of its magnitude and scope.

Researchers will explore how the technology could be applied in future computing as a whole, examining fundamental problems such as spin generation and flow, and data storage, while also developing sample devices. According to the project proposal, the work has the potential to establish Britain as a leading centre for this type of research and “ignite a technology field.”

̽��ֱ��project will be led by Professor Mark Blamire, Head of the Department of Materials Sciences at the ̽��ֱ�� of Cambridge, and Dr Jason Robinson, ̽��ֱ�� Lecturer in Materials Sciences, Fellow of St John’s College, ̽��ֱ�� of Cambridge, and ̽��ֱ�� Research Fellow of the Royal Society. They will work with partners in the ̽��ֱ��’s Cavendish Laboratory (Dr Andrew Ferguson) and at Royal Holloway, London (Professor Matthias Eschrig).

Blamire and Robinson’s core vision of the programme is “to generate a paradigm shift in spin electronics, using recent discoveries about how superconductors can be combined with magnetism.” ̽��ֱ��programme will provide a pathway to making dramatic improvements in computing energy efficiency.

Robinson added: “Many research groups have recognised that superconducting spintronics offer extraordinary potential because they combine the properties of two traditionally incompatible fields to enable ultra-low power digital electronics.”

“However, at the moment, research programmes around the world are individually studying fascinating basic phenomena, rather than looking at developing an overall understanding of what could actually be delivered if all of this was joined up. Our project will aim to establish a closer collaboration between the people doing the basic science, while also developing demonstrator devices that can turn superconducting spintronics into a reality.”

̽��ֱ��initial stages of the five-year project will be exploratory, examining different ways in which spin can be transported and magnetism controlled in a superconducting state. By 2021, however, the team hope that they will have manufactured sample logic and memory devices – the basic components that would be needed to develop a new generation of low-energy computing technologies.

̽��ֱ��project will also report to an advisory board, comprising representatives from several leading technology firms, to ensure an ongoing exchange between the researchers and industry partners capable of taking its results further.

“ ̽��ֱ��programme provides us with an opportunity to take international leadership of this as a technology, as well as in the basic science of studying and improving the interaction between superconductivity and magnetism,” Blamire said. “Once you have grasped the physics behind the operation of a sample device, scaling up from the sort of models that we are aiming to develop is not, in principle, too taxing.”

A Cambridge-led project aiming to develop a new architecture for future computing based on superconducting spintronics - technology designed to increase the energy-efficiency of high-performance computers and data storage - has been announced.

Superconducting spintronics offer extraordinary potential because they combine the properties of two traditionally incompatible fields to enable ultra-low power digital electronics

Jason Robinson

10515 images via Pixabay

Growing quantities of data storage online are driving up the energy costs of high-performance computing and data centres. Superconducting spintronics offer a potential means of significantly increasing their energy-efficiency to resolve this problem.

̽��ֱ��text in this work is licensed under a Creative Commons Attribution 4.0 International License. For image use please see separate credits above.

Yes