Computational Science Community Wiki

Differences between revisions 16 and 18 (spanning 2 versions)
Revision 16 as of 2014-11-27 10:56:02
Size: 5531
Editor: MichaelBane
Revision 18 as of 2014-11-27 11:09:15
Size: 5398
Editor: MichaelBane
Deletions are marked like this. Additions are marked like this.
Line 2: Line 2:
1.30 - 3.30 pm<<BR>>2.220 University Place
Line 4: Line 3:
To book a place at this event please do so through the [[ To Search&returnQs=%3fterm%3dRGPU%26org%3d216 |training catalogue]].

== Agenda ==

 * Introduction<<BR>><<BR>>
 * Michael Bane, Research IT: [[attachment:IT_GpuClub_25Nov2014.pdf|Introduction]]<<BR>><<BR>>
Line 12: Line 7:
      i. Horizon2020
      i. EMiT
      i. Big Data<<BR>><<BR>>
      i. Dave Topping: [[attachment:Topping_GPU_club_November_2014.pdf|Horizon2020 and EMiT conference]]
      i. Malte Vesper: Stream Processing with FPGAs

GPU Club: 25 Nov 2014

Abstracts & speaker biographies are below. Links to slides are as above.

This event was also podcast. (We are making a version available to non-UoM and hope to upload w/c 1 Dec)

The Hartree Centre Energy Efficient Computing Research Programme, by Neil Morgan, STFC

An outline of the Energy Efficient Computing Research Programme (EECRP) at the Hartree Centre. Providing an overview of the systems in place at the Hartree Centre, including the newly procured hardware to support the EECRP including the Maxeler FPGA Data flow system, ClusterVision oil cooled system and forthcoming ARM 64 supercomputer. The opportunities and challenges for the programme, key areas of focus for research and industry engagement, current and planned projects and evolving collaborations.

About the author
Neil Morgan is the Programme Manager for the Energy Efficient Computing Research Programme at the Science and Technology Facilities Council, Hartree Centre. His role includes investigating and developing new architectures and approaches to high performance computing that will support the progression to exascale. Areas of focus include: Application Optimisation, Hybrid Architectures, Energy Aware Software Tools, Resource Management, Standards and Metrics. Neil is an IT professional with a track record of supporting the successful design and implementation of large scale complex programmes. As a proven project and programme manager, he delivers complex change with strong skills in developing governance, strategy and policy to support service improvement. He has significant experience working with stakeholders to translate business requirements into achievable outcomes through business process re-engineering and the innovative application of technology. His specialist skills & interests include High Performance Computing, Information architecture, project and programme management, geographical information systems and spatial data, data analytics and reporting, data modelling (conceptual, logical and physical), taxonomy, metadata, and search, application interface design and usability, pattern recognition and parallel programming.

Towards Real-Time Risk Management on FPGA cards. Evaluation of Risk Sensitivities for the Black-Scholes, Heston and LIBOR Market Model using Maxeler technology, by Grzegorz Kozikowski, UoM

The financial models as Black-Scholes, Heston and LIBOR market model are widely used to price interest rate derivatives and option prices. Of particular interest are the risk sensitivities measured as first-order derivatives of these models rates with respect to all market model parameters. These numerical values known as the Greeks are interpreted and used by traders to neutralize the financial risk and hedge their investment portfolios against price fluctuations on the market. This work investigates a parallel approach to the Greeks' calculation for Black-Scholes, Heston and LIBOR market model. They are evaluated through a single Monte-Carlo simulation by the Adjoint method. The Monte-Carlo implementation is supported by FPGA card (Maxeler framework) and compared to a sequential version executed on Intel Sandybridge machine. The combination of the Adjoint and FPGA has improved performance by three orders of magnitude and contributes to better accuracy if compared to pathwise, finite difference or likelihood methods. The work present energy efficiency comparison of FPGA against CPUs.

About the author
Grzegorz Kozikowski is a PhD Student of Computer Science at University of Manchester since 2013. His academic career concerns High-Performance Computing and Numerical Methods with application in Finance. Based on Bachelor and Master Thesis two publications were published: Automatic Differentiation and Interval Arithmetic using OpenCL in PARA Proceedings 2012 and Parallel Approach to Monte Carlo Simulation for Option Price Sensitivities Using the Adjoint and Interval Analysis in PPAM Proceedings 2013. In 2009 Grzegorz was working as a Junior Consultant at IBM Global Business Services. In 2011 he joined IBM Software Group and worked until 2012, where he was responsible for software development using High Performance Computing (GPU, multicore architectures). His current research investigates application of HPC architectures as (GPU, FPGA, Xeon-Phi) to hedging derivatives, interest rate derivatives and option pricing. His interests concern High Performance Computing, Monte-Carlo simulations, Software Development and Derivatives.