ARSC T3D Users' Newsletter 112, November 15, 1996

T3E Teraflop Machine Publically Announced by SGI/Cray


> CRAY UNVEILS TERAFLOP SUPERCOMPUTER
> Silicon Graphic's Cray Research unit has unveiled its new CRAY T3E-900
> supercomputer capable of performing a trillion calculations per second,
> becoming the first company to bring a teraflop system to market.
> Shipments will begin in Spring 1997, with prices starting at $500,000.
> Cray anticipates selling the super-fast systems to oil companies and
> other exploratory concerns, which could use the machine's power to map
> and analyze sites to find oil and minerals for extracting.  The new
> machines can handle in just a few days calculations that it would take
> less powerful systems up to three months to process.
> 
> (New York Times 12 Nov 96 A18)
For more information, see:
http://www.cray.com/PUBLIC/WHATS_NEW/PRODUCTS/CRAY_T3E-900.html

Here's an excerpt which explains where the teraflops number comes from:


> Demonstrated Sustained TFLOPS 
> =============================
> 
> The CRAY T3E-900 system is the first computer system capable of
> demonstrating capabilities of a teraflops (trillion calculations per
> second) sustained performance. On the LINPACK benchmark, a software
> program that tests a computer's sustained performance, a CRAY T3E
> system registered 504 megaflops per processor.  According to Cray
> benchmark specialists, based on this data a fully configured CRAY
> T3E-900 system will deliver in excess of a teraflops (trillion
> calculations per second) of sustained performance on the LINPACK test.
> 
> Cray is also the only company to publish NAS parallel benchmark results
> for large parallel system configurations, demonstrating the ability to
> scale to 100s and up to 1000s of processors. Cray has already received
> orders for CRAY T3E systems with both 512 processors and 1024
> processors. Competing systems don't scale beyond 128 processors.

EPCC T3D Documentation

Thanks to Stuart Paton of EPCC who sent the URL:

http://www.epcc.ed.ac.uk/t3d/documents/techreports.html

It takes you to a list of technical reports on the T3D. I've reproduced the titles and synopses:

T3D Technical Reports

Porting PVM codes to the T3D
This document describes in detail the steps needed to port a PVM code to run on the T3D, and offers some advice about possible pitfalls.
Craft Performance Optimisation
This document describes various methods for increasing the speed of codes written in Craft, the Data Parallel programming style on the Cray T3D.
HPF on the Cray T3D: A Comparison with Craft
This document describes a performance comparison between codes written in Craft, the Data Parallel programming style on the Cray T3D, and the Portland Group implementation of HPF.
Fast Fourier Transforms on the Cray T3D
This report benchmarks the Cray library routine PCCFFT3D for performing parallel complex-to-complex FFT's on three-dimensional distributed data sets.
Optimisation of QCD simulations
This report documents work done at EPCC on the UKQCD simulation codes. This material formed he basis of a talk at the following conference. "Meeting on the Optimization of Codes for CRAY MPP Systems" January 24-26, 1996 Pittsburgh, Pennsylvania
Parallel Random Number Generators
This report surveys current methods of high quality random number generation and investigates parallel implementations of these algorithms.
IO on the Cray T3D
This IO Technical Report discusses why IO is important, how the I O works on the T3D, parallel IO and some Cray extensions. An IO example program is tested using methods discussed and performance figures are given. This report is based on several IO chapters that were included at the end of the T3D Optimisation course.

Quick-Tip Q & A


A: {{ You discover a suspicious file called "a.out". You think it's an
      executable. How can you determine which platform it was targeted 
      to run on? }}
      
   # 'mppsize' utility (Thanks to Barbara Herron of LLNL):
   # Run on any file: if it's a T3D executable, it returns lots of 
   # information on the file; if not, it returns a mild warning.

   denali$ mppsize a.out.YMP
   This doesn't look like a CRAY-T3D absolute to me.

   denali$ mppsize a.out.T3D
   Startup program is:     #!/mpp/bin/mppexec
    [ ... lots of other good info including:
          npes, fixed vs plastic, and summary of load map ...  ]
      
   # 'file' utility (Thanks to Stuart Paton of EPCC)
   # This utility performs a series of checks on a file to try to
   # discover its contents. 

   kelvin$file a.out 
   a.out:          MPP absolute

   kelvin$file a.out 
   a.out:          executable J90 cache safe CRAY Y-MP not stripped

   kelvin$file hello.T 
   hello.T:        ASCII Compiler Information File - version 2 - C

   kelvin$file app.rif 
   app.rif:        data

   kelvin$file hello.c 
   hello.c:        c program text

   kelvin$file letter 
   letter:         ascii text

   etc.


Q: Do programmers get carpal tunnel syndrome?  How bad is it?  Can you
   avoid it?

[ Answers, questions, and tips graciously accepted. ]


Current Editors:
Ed Kornkven ARSC HPC Specialist ph: 907-450-8669
Kate Hedstrom ARSC Oceanographic Specialist ph: 907-450-8678
Arctic Region Supercomputing Center
University of Alaska Fairbanks
PO Box 756020
Fairbanks AK 99775-6020
E-mail Subscriptions: Archives:
    Back issues of the ASCII e-mail edition of the ARSC T3D/T3E/HPC Users' Newsletter are available by request. Please contact the editors.
Back to Top