Improving the energy efficiency of computation and data analysis
24 Nov 2014
The session's first meeting of the University’s GPU Club will be at 2pm on Tuesday 25 November in Room 2.220, University Place
Neil Morgan, STFC, will outline how the Energy Efficient Computing Research Programme at the Hartree Centre will investigate new hardware architectures and develop new approaches to reducing the energy footprint of computing and data analytics.
Greg Kozikowski, of Computer Science at The University of Manchester, will outline the relative computational and energy performances of GPUs and FPGAs for modelling risk sensitivities of financial models such as those used on the Stock Exchange.
There is the opportunity for attendees to have a “pop up” session of up to four minutes with up to two slides to highlight their work on emerging technologies (GPUs, FPGAs, Xeon Phi and so on) and on energy efficient computing.
The talks begin at 2pm sharp, finishing about 3.30pm. There is also an opportunity to network over a buffet lunch from 1.30pm.
Abstracts and further details are available at:
To attend, please register at:
For further information or enquiries please contact:
In recent years, researchers have made use of GPUs (Graphical Processing Units) and Xeon Phi. These are both, currently, PCIe cards that can be used to accelerate kernels of code. They are particularly efficient in performing numerous vector-like operations on the same set of data: more efficient that a generic CPU and thus faster. The Field Programmable Gate Array (FPGA) approach comprises a high level design of required logic units to solve a given problem and “reconfiguring” the Gate Arrays to match this design. Again, this approach can give substantial speed up over a traditional CPU for specific problems.
The global environmental cost of computing is comparable to that of aviation, according for approximately 2% of total Carbon emissions (Gartner, 2007). At The University of Manchester, it is estimated that IT accounts for approximately 12% of Carbon emissions (CMP, 2009). It is therefore imperative that to investigate methods to perform model simulation and data analysis using less energy. This emerging field, commonly labelled “energy efficient compute”, has been considering how to measure the energy costs, and the role of accelerators such as GPUs, Xeon Phi and FPGAs.