Developing

1
Developing Deploying Advanced Astrophysical
Algorithms on Novel Supercomputing Hardware

• PI Robert J. Brunner
• bigdog_at_uiuc.edu
• Project Summary
• We are implementing point statistics and
  machine learning algorithms on novel
  supercomputing hardware to ascertain the efficacy
  of these platforms for forthcoming petascale data
  analyses. These systems range from high
  performance reconfigurable computers that
  leverage FPGAs to Cell-based systems, and
  graphic-chip based vector processors.
• Plans for Upcoming Year
• Complete optimized 2-point correlation code port
  to HPRC platforms
• Initiate port of power spectrum code to SRC-6
  platform
• Initiate 2-point correlation code port to
  alternative platforms

• FY06 Highlights
• Invited Presentation on Cosmological Applications
  of HPRC at RSSI 2006 conference
• Evaluated SRC-6 MAPC and MAPE reconfigurable
  processors
• Evaluated SGI RASC Athena and RC100 compute
  blades
• Initiated several new algorithm developments
• Several publications in progress
• Graduate Student trained at SRC headquarters on
  using SRC-6 MAP station

Applications Completed development of brute force
two-point angular correlation measurement
(initial benchmark results shown in included
figure) on both an RC100 system (10x performance
improvement over dual 1.4 GHz Intel Itanium 2
system) and on single and dual SRC-6 MAPE systems
(90x speedup over 2.8 GHz Intel Xeon system).
http//lcdm.ncsa.uiuc.edu
AISR 2006-09
HPRC, FPGA, Correlations