UK world leadership in many fields has HPC at the core
From the world of Formula 1 through jet engine design to next generation drug discovery, the UK is a leader in high value, high skill industry sectors. While they all began by using the intuition of skilled practitioners – science and engineering as craft – High Performance Compute (HPC) is now central to continued success in these fields.
The evolution of both the capacity and the applications of HPC has been continuous and developed from the very early days of electronic computation. The use of Colossus to decrypt coded enemy communications at Bletchley Park is well documented, and in the 1950s, Cambridge University Mathematical Laboratory’s EDSAC machine was used for some of the first computational chemistry investigations. Collaborating with Manchester University, Ferranti developed new supercomputing capabilities, supplying the Met Office with their first computer. As new techniques in computer science, electronic engineering and materials science were developed, new application spaces and higher performance opened. For several decades, supercomputers were a breed apart from general purpose machines, dedicated instead to scientific computing with vector pathways and fast floating-point maths.
General purpose computing
As the computing industry expanded and computers became commodity items, it was realised that the volume of general-purpose server shipments was creating economies of scale that were funding significant research and development in microprocessor design. While not optimised for HPC, the cost of generic servers was so low that centres could buy enough so that for certain classes of application the performance attained by a “Beowulf Cluster” was as good or better than a traditional supercomputer.
Over the past few decades, commodity silicon has dominated the Top500 list of the world’s fastest supercomputers. A symbiosis between the HPC and general-purpose markets has led to features in server silicon being introduced for scientific computing that were then employed more widely, while the size of the combined market has allowed the development costs to be amortised over a much larger set of users.
In recent years, the use of GPUs to accelerate the processing required for AI/ML techniques has become established, as engineers realised that the matrix mathematics required for graphics is shared by deep learning techniques, and that the hardware inside GPUs could be repurposed for other uses. Primed by the gaming industry, GPUs now provide energy efficient acceleration to a significant proportion of leading supercomputers.
The ability to train AI networks at this scale has led to another revolution in HPC, with models now becoming key to advances in fields as diverse as astronomy, medical imaging and particle physics.
Enter the cloud
Another revolution is happening now. The cloud has revolutionized computing – providing compute on-demand.
The first cloud computing services were also built from commodity hardware. Here, some HPC applications like Electronic Design Automation (EDA) in the semiconductor industry, or – another high-value sector for the UK – bioinformatics, have found a good fit.
Arm itself is a significant consumer of compute cycles for the EDA workloads that it must run to validate its designs – and while we still run our own large-scale compute in our datacentres, like many organisations we take a hybrid approach and much of our compute is now run effectively in the cloud.
Yet again, economies of scale outside of HPC is setting up another round of innovation for HPC.
Arm is now one of the key enablers for cloud providers to build bespoke hardware for the tasks ahead, offering advanced IP for processing and computing subsystem design. Leveraging the economies of scale provided by the smartphone market to support the creation of a new generation of processors applicable to cloud and HPC markets, Arm is now developing its next designs using cloud compute powered by Arm-based processors.
In fields such as weather forecasting or aerodynamic modelling, simulations are so large that they must split over many – indeed hundreds or thousands of – servers, all computing their section of the simulation and synchronizing the partial calculations throughout. This closely coupled compute needs very fast connections (“interconnects”) between the servers that process the workload.
Some cloud providers have added commercial supercomputers into their datacentre, or supercomputer technology in the form of high-end interconnects. Others have developed their own high-end interconnects, enabling hosted services to grow the scope of HPC applications that can be run.
Evolution continues
The innovation in architectures is continuing across the stack. The Wafer-Scale Engine from Cerebras is an example of innovative thinking about how modern fabrication techniques can be adapted to the training problems of today. The developments in quantum computing from Google, IBM, Microsoft and others promise to revolutionise areas that previously could not be feasibly simulated, at the same time posing challenges to the way that we think about encryption today.
The UK’s university base has been central to the story of HPC. From the early days of Baby, EDSAC and the Manchester Mk1, it continues to research new approaches to HPC. Teams at institutions like Bristol and Edinburgh are working on the next generation of platforms in partnership with other academic centres and industry, both as suppliers and users.
While modern networks mean that compute can be accessed across oceans, the geographical location of computing centres still matters. There are concerns for energy costs associated with moving large datasets and the speed of light is an important consideration. However, some of the most pressing issues come from geopolitical requirements. Personal medical data often needs to remain within country, and export control restrictions may apply to the design of dual use components – such as microprocessors.
Continuing to build a strong provider ecosystem in the UK is critical for the country’s ability to maintain a leading position in research and development of the next generation of technologies – in aerospace, life sciences, electronics and academic research.
Laura Foster
Laura is techUK’s Head of Programme for Technology and Innovation.
Rory Daniels
Rory joined techUK in June 2023 after three years in the Civil Service on its Fast Stream leadership development programme.
Elis Thomas
Elis joined techUK in December 2023 as a Programme Manager for Tech and Innovation, focusing on AI, Semiconductors and Digital ID.
