Itanium Lab
Applied Mathematics Electrical and Computer Engineering University of Colorado at Boulder

	Overview

The Itanium Lab is a cluster of nine high performance processors dedicated to large scientific computation. The Applied Mathematics Department and the Electrical and Computer Engineering Department of the University of Coloarado at Boulder use these machines for research in computational fluid dynamics, geophysical and astrophysical fluid simulations, turbulence, chaotic systems, nonlinear dynamics, compiler efficiency, and other areas. The lab is the result of a collaboration between the CU Applied Mathematics Department, CU Electrical and Computer Engineering Department and Hewlett-Packard, who is commited to supporting research and education. The principle investigators are professor Keith Julien in Applied Mathematics and professor Dan Connors in Electrical and Computer Engineering.

	Hardware

• Five single processor HP zx2000 Workstations
  • Processor: Single Intel Itanium 2 900MHz processor (32KB L1 cache, 256 KB L2 cache, 1.5MB on-chip L3 cache)
  • Main Memory: 4 GB DDR SDRAM
  • Hard Disk Drives: Two 73GB 15K Ultra320 SCSI HDD per workstation
  • Standard Performance Evaluation: integer floating point integer rates floating point rates
  
  
• One four processor HP Integrity rx5670
  • Processors: Four Intel Itanium 2 1.3GHz processors (32KB L1 cache, 256 KB L2 cache, 3MB L3 cache)
  • Main Memory: 24 GB DDR SDRAM total shared memory
  • Hard Disk Drives: Two 73GB 15K Ultra320 SCSI
  • Standard Performance Evaluation: integer floating point integer rates: 1 2 4 floating point rates: 1 2 4
  
  
• Gigabit Network
  • Switch: Netgear GS108 8-port Gigabit switch
  
  
• Operating System: Redhat Linux

	Software

	Progress & Results

• Current users:
  • 3D Eddy Current Problem Eunjung Lee and Tom Manteuffel are working on the optimal convergence of the Eddy current system in 3D.
    
    
  • Numerical Plasma Simulation with VORPALViktor Przebinda and Dan Connors are using VORPAL to study the Itanium architecture and Intel's C++ compiler.
    
    
  • Atmospheric Stratified Shear Flows Matt Tearle, with Joe Werne and Keith Julien, is using Optimal Perturbation Analysis to investigate the transition from laminar to turbulent flow.
    
    
  • Oceanic energy cascade Mark Petersen, with Keith Julien and Jeff Weiss, is studing the energy cascade and vortex characteristics in an intermediate scale quasi-geostrophic model.
    Recent Results: timing and size projections, diagnostics plots, and three dimensional volume rendering
    
    
  • Quasigeostrophic Ellipsoidal Vortices Neil Burrel, Keith Julien and Jeff Weiss are studying three dimensional vortex interaction in quasi-geostrophic (oceanic) flow.
    
    
  • 3D Fluid Mixing Paul Mullowney, with Keith Julien and James Meiss, is using dynamical systems tools to study three dimensional fluid mixing. His code calculates Lyapunov exponents to measure the amount of chaos and mixing(dispersion) in the system.
    
    
  • Convective plumes Michael Sprague, with Keith Julien, Edgar Knobloch (UC Berkeley), Ralph Milliff (CORA), and Joe Werne (CORA) are studying convective plumes in fluids which are rotating, stratified, and heated.
    
    
  • Blood Flow in Compliant Vessels Jeff Heys, with Tom Manteuffel, Steve McCormick, a group of pediatric cardiologists, and others, is developing a mathematical model and a software product to simulate flow in the cardiovascular system. ParaFOS uses a new First-Order System Least Squares (FOSLS) approach to solve simultaneously for blood flow and wall deformation.
  
  
• Itanium Benchmarking
  • In October 2003 we collaborated with Steven G. Johnson and Matteo Frigo of MIT to benchmark numerous FFTs on the Itanium 2 processor. Results for our machines have been released on their BenchFFT site and include a comparison of 26 FFT packages using 24 separate tests. This was the first FFT comparison conducted on Itanium 2 machines by this group.
  
  
• NAG Compiler Testing
  • The NAG Fortran compiler causes a "Segmentation Fault" at the end of program execution on these Itanium 2 processors. In October 2003 we assisted the NAG technical staff in diagnosing the error by running test codes. These were some of the first tests of the NAG compiler on the Itanium 2 architecure. The problem can be avoided by using the -unsharedf95 compilation flag.
  
  
• Intel Fortran Compiler 8.0 Testing
  • The Intel Fortran Compiler version 8.0 causes a "Segmentation Fault" when automatic arrays larger than ~10MB are declared. This problem does not occur in version 7.0, but can be corrected by typing limit stacksize unlimited at the unix prompt (in c-shell). We contacted the Intel compiler group about this problem.

	Contacts

• General questions on our Itanium Lab, including collaboration requests:
  • Michael Sprague, Michael(dot)Sprague(at)colorado(dot)edu

Last updated 3/1/04 by M. Petersen