EPCC, the supercomputing centre at the University of Edinburgh, is perhaps best known for being home to the UK’s most powerful supercomputer, ARCHER2.
Sitting alongside it, very soon, will be our first Graphcore IPU Pod.
These two quite different machines perfectly illustrate the centre’s mission: to accelerate the effective exploitation of novel computing throughout industry, academia and commerce.
In Graphcore’s Intelligence Processing Unit, we recognised a tool that will be of huge interest to our many partners, allowing them and the EPCC team to explore innovative applications of AI within their scientific and business challenges.
The IPU’s unique architecture is rooted in a deep understanding of the novel and idiosyncratic compute requirements of AI. Not just today’s workloads, but those dictated by the field's general direction of travel. For example, the performance of the latest IPU’s 1,472 totally independent MIMD processing cores for efficient sparse computation is really impressive. Multiply those across a large IPU Pod system and the level of fine-grained AI compute reaches another level.
The sheer processing power of the Bow IPU – more than 22 PetaFLOPS of AI compute across a Bow POD64 system – means that the often iterative process of discovery can be dramatically accelerated, moving us forward further and faster.
Data driven innovation
EPCC’s first Bow Pod system will reside within the Edinburgh International Data Facility (EIDF), a set of computing services that underpin the UK and Scottish Government-funded Data Driven Innovation (DDI) Programme in Edinburgh and SE Scotland.
Established in 2018, DDI aims to support innovation, skills, job creation and economic growth across a range of industries, including financial services, agritech, health and social care, and the creative industries, as well as in the public sector and academia.
The EIDF is a large multi-faceted Private Cloud infrastructure that has at its heart two Data Science infrastructures: one is an open infrastructure focussed on applications of AI and data science arising from all of the projects that form the Data Driven Innovation Programme. The other infrastructure focusses on the provision of a Private Cloud for personal data – a key task is hosting the Scottish National Safe Haven.
Our Bow Pod system will initially sit in the open data science infrastructure, allowing a broad range of use cases to be brought to it. In the future we will also look to adding a system inside the Safe Haven part of the infrastructure.
We see this initial investment as the start of a long-term relationship with Graphcore and we look forward to bringing a huge variety of different projects to the system over the coming months.
We are also excited by Graphcore’s product roadmap – including the Exascale Good Computer. As home to one of the world’s most powerful HPC systems, we are already giving careful consideration to the sort of large-scale compute needed to support the continued evolution of AI and the Good Computer looks very interesting.
In 2020, EPCC awarded a £100m contract to HPE, initially for 5 years, to provide the IT infrastructure components for the EIDF.
We have worked closely with HPE on this project and throughout their journey to becoming a Graphcore Elite Partner, adding the IPU to their customer offering.
We are grateful to both HPE and Graphcore for their support throughout this project and look forward to building our IPU compute capability together in future.