7 Keble Rd
Oxford, Oxfordshire OX1 3QG
UK
Dr Paul Richmond
University of Sheffield
June 15, 2016 – 13:00 to 14:00
Conference Room
Oxford e-Research Centre, 7 Keble Road, Oxford, OX1 3QG
7 Keble Road
Oxford
OX1 3QG
No booking required Open to all Many-Core Series Sandwich Lunch provided
Demand on transport services and infrastructure is a growing problem around the globe, for instance traffic demand is projected to increase by up to 55% in the UK by 2040. Existing infrastructure is often incapable of meeting current demand resulting in congestion, overcrowding and disruption to travellers.
Predictive simulations are urgently required to help design transport management systems and to increase the capacity and effectiveness of current infrastructure. This talk will describe ongoing work with commercial partners in the transport sector to accelerate large scale network flow simulations. This work aims to allow increasingly large scenarios to be explored within reasonable time constraints allowing the simulation of national infrastructure. More specifically, this talk will introduce the network flow simulation techniques and describe in detail the process and lessons learned from optimising the algorithms for Maxwell generation NVIDIA GPUs. The techniques described use state of the art graph traversal techniques and block level load balancing. Performance of up to 40x is reported compared to an optimised serial version and over 6x against a 16-core multi-core version.
About the speaker:
Dr Richmond is a research focused Research Software Engineer who has recently been awarded one of only six EPSRC Early Career Research Software Engineering (RSE) Fellowships. The focus of this fellowship is in facilitating the use of accelerated architectures such as Graphics Processing Units (GPUs) to accelerate scientific discovery. He is developing software techniques, a provision of skills and training material and building a community to help drive the use of accelerators into mainstream science and engineering.
Dr Richmond has a proven track-record of forming inter-disciplinary collaborations to achieve agenda-driven research. His work focuses on developing software which facilitates the pioneering use of emerging high-performance computing architectures for complex systems simulation within computational science and engineering. He is an excellent communicator with a long term record of engaging scientists and engineers from diverse interdisciplinary fields to deliver requirement-driven software solutions.
– See more at: http://www.oerc.ox.ac.uk/events/Paul_Richmond