Accelerating the deep learning research with AI supercomputer
As part of the new Sheffield Advanced Research Computer cluster the new supercomputer plays an important part in the department’s plans to become a centre of expertise for deep learning in the UK. The cluster is equipped with other NVIDIA Tesla graphic processors enabled nodes to meet the increasing computational demands of research is data-rich areas of machine learning and simulation.
Existing research groups such as the AVCOGHEAR project have reported large increases in speed when training their complex models. This EPSRC-funded project applies machine learning in the development of next-generation audio-visual hearing aids led by Dr Jon Barker.
Professor Lucia Specia who heads the MultiMT project aims to develop new machine translation methods that incorporate multimodal information for learning and inference, of the new facilities Professor Specia said, “The extra processing power and memory bandwidth available in the DGX-1 makes the computation extremely fast and efficient. This makes the extensive experiments for the training of Deep Neural Networks which are needed for this project feasible“.
The DGX-1 is also being used for accelerated computing, where it is being applied to speed up the simulation of national transportation systems in projects with both ATKINS and the Department for Transport.
“The NVIDIA DGX-1 AI supercomputer is helping researchers at the University of Sheffield accelerate their expertise in deep learning techniques and undertake AI research in ways not yet explored,” said Alex White, Vice President of Enterprise for Europe, Middle East and Africa at NVIDIA.
“We expect the results of the researchers will not only benefit the university, but also the citizens of the United Kingdom and society at large.”
With the help of newly formed Research Software Engineering in Sheffield (RSES) group led by Dr. Paul Richmond and Dr. Mike Croucher, the department intends to provide research and industrial consultancy as well as training in order to apply deep learning expertise to a wider range of research especially outside the field of computing.