Earlier this month, the UK Government announced a £300 million package of support for the UK’s AI Research Resource, strengthened by a further commitment of £500M in the Autumn Statement. Included in these announcements was funding for a new Cambridge supercomputer – ‘Dawn’ – being developed by a partnership between the University of Cambridge, Dell, Intel, and StackHPC. This next-generation High Performance Computing and AI platform offers the opportunity to accelerate AI research in the UK, helping researchers to build world-leading AI systems that deliver scientific and societal benefit.
The Dawn system will significantly expand Cambridge’s compute capabilities, supercharging AI and simulation research across the University. With five times the performance of Cambridge’s previous GPU system, it is designed to meet the needs of researchers who require high performance computing for AI, with implications for our ability to use AI for research across climate science, astronomy, healthcare, and more. To translate this capability into tools and applications that deliver real-world benefits, the system needs to be accessible to a wide range of research users across the AI and simulation communities.
Over the coming months, the Dawn team will be developing points of access, working processes, and support packages to allow researchers from across the Cambridge community to make use of this resource. To help design these interfaces, we’re inviting a cohort of pioneer projects with High Performance Computing needs to have early access to the Dawn system. This joint call from ai@cam, Research Computing Services, the Cambridge Open Zettascale Lab, the Cambridge Centre for Data Driven Discovery, and the Institute of Computing for Climate Science will offer these pioneer projects early access to Dawn and support using the new system. Lessons from these interactions will feed into the design of support services built around this new compute facility.
“This is a fantastic research resource for Cambridge’s AI & simulation communities” said Paul Calleja, introducing the call, “as we shift to operationalisation, we want to develop a good understanding of the AI applications, workflows and user environment needed by the broad cross section of the AI and simulation communities. We hope these pioneer projects will help us understand the different needs that arise across the research community, and to tailor our system environment accordingly.”
The call for pioneer projects is open to researchers across the University of Cambridge. To propose a pioneer project, please fill in the form at the link below by 17:00 on 8 January 2024. For further details, please check out the FAQs or email us at support@hpc.cam.ac.uk. After 8 January, we’ll select a group of projects representing different research challenges and user needs to be supported in a cohort of pioneer projects.
Link to apply: https://forms.gle/futP2GvjWmnKLaFS9
If you’d like to find out more about the capabilities of the Dawn system or ask questions about the pioneer project initiative, you can join Q&A session on 18 December at 16:00 - 17:00. To register for this session, please use the following link: https://cam-ac-uk.zoom.us/meeting/register/tZAoc-mqpzwtHNTV2n5hLJ2J3r3cxx3E3-Ci or alternatively email your questions to support@hpc.cam.ac.uk
Frequently asked questions
Why have you launched this call?
This new compute resource will support AI and simulation research across the UK and the University of Cambridge. To make sure researchers across the community can capitalise on the benefits Dawn offers, we will need well-designed interfaces and working processes that make the system easy to use and understand. We hope that working with this cohort of pioneer projects will help us tailor our ways of working to the needs of the AI and simulation communities. In return, projects will have early access to the new compute resource and associated support, such as software engineering expertise.
Who can apply?
Applications are open to any postgraduate student or member of staff at the University of Cambridge, working in any discipline, including natural, physical, social, medical, and computer sciences, arts, humanities, and engineering.
What kind of projects will be supported?
Dawn is being designed to support converged AI and simulation workloads. To help us understand different types of stakeholder needs, we’re looking to engage with a variety of pioneer projects in the AI and simulation communities. You might already have extensive experience of engaging with high performance computing facilities, or none at all. You might be working on simulations specifically, or broader applications of AI. We’re interested in projects both with computationally heavy needs, and use cases with more routine compute requirements. We can provide support with the software engineering required to make use of the Dawn system, but we do ask that your team has sufficient expertise in AI to develop your project ideas. You could be working across a range of disciplines. What pioneer projects will share is a need for High Performance Computing and a willingness to give some time to discussions with the Research Computing Services team to help design their workflows and develop use cases for the system.
What support will I get?
In addition to early access to the Dawn system, we’re offering pioneer participants support in the form of research software engineering advice, and access to services such as the Accelerate Science Machine Learning Engineering Clinic. In return, we’ll ask for your support translating what we learn from setting you up with Dawn into case studies or user guides, to help others in the community interact with the system.This first cohort of projects will be able to access the Dawn system free of charge. Future pricing systems for access are currently being developed.
What are the capabilities of the Dawn system?
The Cambridge Open Zettascale Lab is hosting Dawn, the UK’s fastest artificial intelligence (AI) supercomputer, which has been built by the University of Cambridge’s Research Computing Services, Intel and Dell Technologies. Once set up, Dawn will provide 256 Dell XE9640 servers nodes, each server has 4 Intel Data Centre Max 1550 GPUs (each GPU has 128 GB HBM RAM configured in a 4-way SMP mode with XE-LINK). Each server has 2 Gen 5 XEONs, and 1TB RAM and 4 HDR200 Infiniband links connected to a fully non-blocking fat tree. In total there are 1024 Intel GPUs. The system is an Intel GPU system, not NVIDIA, thus your application will need to be written with oneAPI (not CUDA) for Intel based GPUs or use standard AI libraries that have been ported to the Intel GPU. You can read more about the system here: https://www.hpc.cam.ac.uk/d-w-n You can find out more about the Cambridge Open Zettascale Lab here: https://www.zettascale.hpc.cam.ac.uk/
How will my proposal be evaluated?
This is a trial initiative. If we receive a large volume of applications, we’ll shortlist to ensure we bring a diverse range of disciplines, levels of familiarity with HPC systems, and needs into our pioneer cohort.
Over what timescales will pioneer projects operate?
The Dawn system will be coming online in January 2024, and we anticipate working with projects to set them up with the service in the first half of 2024.
What information is required for the application?
- Name
- Department
- Project title
- Project summary (max. 1000 characters)
- Summary of compute needs (if you know them) (max. 1500 characters). If you know these details, it would be helpful if you could list the applications and or libraries you are using and let us know if there is a oneAPI Intel GPUport available.
- Level of familiarity with high performance compute systems (max. 1000 characters)
- Type of support you might need to make use of the Dawn system (for example, software engineering support) (max. 1000 characters)
- Readiness (timescale for deployment) (max. 1000 characters)
How can I apply?
To apply, please submit an application form at https://forms.gle/futP2GvjWmnKLaFS9 by 17:00 on Monday 8 January 2024.
