http://www.cardiff.ac.uk/arcca/services/events/HPCSIG-NovelArchitecture.html
HPC-SIG: Novel Architecture Focus Group
Cardiff University, 25th April 2008
A meeting of the Novel Architecture Focus Group of the HPC-SIG will
be held in the CUBRIC Seminar
Room, School of Psychology, Cardiff University commencing 10:45am
on 25th April 2008.
The
meeting will cover presentations from Novel Architecture suppliers in
the UK who provide technical overviews of their offerings and
demonstrations of their deployment. These presentations will be driven
by application case studies, with user participation in the session,
designed to illustrate the effort involved in the development of new
codes and migration of existing software focusing throughout on the
real performance gains achieved.
Companies who will be presenting at the meeting include nVIDIA,
Clearspeed, Bull and SGI.
Workshop Agenda:
| Time |
Agenda |
| 10:45 - 11:15 |
Registration (plus
refreshments) |
| 11:15 - 11:30 |
|
| 11:30 - 12:30 |
Mike Giles (Oxford University)
& JC Baratault (nVIDIA)
|
| 12:30 - 13:30 |
Marc Mendez - Bull (CUDA)
|
| 13:30 - 14:00 |
Buffet Lunch
|
| 14:00 - 15:00 |
Michael Woodacre - SGI (FPGA)
|
| 15:00 - 16:00 |
Simon McIntosh-Smith - ClearSpeed
& Philip Brown (Bristol University) Presentation to follow
|
| 16:00 - 16:30 |
Nick Avis - Cardiff: Reconfigurable
Computing |
| 16:30 - 17:00 |
Panel Discussion / Tour of Machine
room. |
| 17:00 |
Close
|
Presenter and Presentation Abstract:
| Company |
Presenter |
Title & Abstract |
| (CUDA) |
Marc Mendez |
Accelerating HPC applications.
While
accelerators look like a good idea to add both space and power
low-budget processing units to HPC systems, it is foreseen that their
integration will require some adjustments so they are made usable to
every scientific user:
- development kits have to improve;
- resource management may not handle their peculiar
properties;
- most of all, wild parallelism proposed require new
algorithms and thoughts.
Closing up with just announced and future systems aims at
opening to further discussions and moving the subject to new scales.
|
| ClearSpeed |
Simon McIntosh-Smith (ClearSpeed)
Philip Brown
(Bristol University) |
ClearSpeed Acceleration Technology for HPC: From
Theory to Practice.
With
the advent of accelerators designed specifically for speeding up 64-bit
floating point scientific computing codes, whole new classes of
applications can now beneit from the greater performance and increased
compute density offered by these solutions. In this presentation we
will describe ClearSpeed's accelerator architecture, programming model
and product family. In addition we will review the results of three
different application ports: one port by ClearSpeed and two by our
users.
These application ports will cover:
- The port of the Density Functional Theory component of the
Molpro Quantum Chemistry code by Dr. Fred Manby's team;
- The port of the BUDE protein-ligand docking code by Dr.
Richard Sessions' team;
- The port of the AMBER molecular dynamics code by
ClearSpeed staff.
The
description of the Molpro port will be presented by the developer who
has been porting the code, Philip Brown from Bristol University.
|
| Oxford University |
Mike Giles |
GPUs: The next big thing in HPC?
In
this presentation, I will discuss the very significant potential of
GPUs, offering at least 10x better price/performance and energy
efficiency than conventional Intel/AMD based systems. Based on my
experience with implementing Monte Carlo and finite difference solvers
on nVIDIA GPU's using their CUDA development environment, I will also
discuss the programming effort required to achieve this potential, and
offer my thoughts on what is needed for GPUs to become a mainstream HPC
solution. |
| nVIDIA |
JC Baratault |
nVIDIA CUDA software and GPU Parallel
Computing Architecture.
|
| SGI (FPGA) |
Michael Woodacre |
Application acceleration and the use of
FPGA Technology.
This talk will present an overview of technology available for
application acceleration, and provide as an example the development and
results obtained accelerating the bioinformatics code, BLAST-N.
|
|
