Re: WTB Math coprocessor

[achin@fiu.edu (Albert Chin-A-Young)]

Someone@Somewhere (eaeu899@rigel.oac.uci.edu) wrote:
: does any one know where i can get a math coprocessor for the iigs? i knew 
: there was one a long, long time ago. does it still exist?

Dirk Froehling posted this awhile back:
Here is some info about the already available NumberCruncher floating
point card:

NumberCruncher features:

  - totally compatible with the FPE from Innovative Systems
  
  - much less sensible to heat, voltage problems etc.
  
  - supports the FPE SANE patch for speeding up any program that does
    floating point calculations
    
  - you can compile ORCA/Pascal and C programs to use the it directly
    with the special floatlib for the FPE provided by ByteWorks
  
  - works in any slot without need for setting it to "Your Card"

The current price of a NumberCruncher card without 68881 is DM 250,- plus
shipping. If you already have a FPE with the copro socketed, you can use it.
You can use any 68881 or 882 in a ceramic package that runs at least 10 Mhz.
The NC runs at this speed, if I recall it correctly (more is not necessary
since the slot I/O is the bottleneck).
I think the price for NC+881 is DM 480,-. You are probably better off
getting the 68881 by yourself. The current exchange rate is about
1$ = 1.8 DM I think, so DM250 will be $139.

For orders or exact shipping info please contact

Andreas Schmidt
Ruesingstr. 59
D-44894 Bochum
Germany

Phone:    (011 49) 234 9230 833
FAX:      (011 49) 234 9230 846
Internet: as@mez.ruhr-uni-bochum.de


Here are some Benchmarks using ORCA/Pascal's direct 68881 support for
the NumberCruncher. The MHz column means at what speed a plain IIGS has
to run to get the test done in the same time:
 

The Tests:                        MHz    Seconds    Ticks

3 * 1899 Primes at              12.67       0.25       15
10 * Fibonacci at                9.46       7.80      468
Quicksort: 1000 Strings at       8.39       1.55       93
1000 Iterations of Float at     55.95       1.62       97
15000 Gamm Units at             24.06       2.62      157
Savage at                      560.56       0.30       18
Integer Math at                 11.74       8.47      508


Primes: Calculates the first 1899 prime numbers with the sieve of
Eratosthenes.

Fibonacci: Recursively computes Fibonacci numbers to test the speed
of procedure calls.

QuickSort: Sort alphabetically 1000 10 character strings.

Float: Test simple floating point operations.

Gamm: Test the speed of floating point operations in a mix that is
typical of scientific and engineering applications.

Savage: Test the speed (and stability) of floating point functions
like arctan, exp, ln, sqrt.

Integer Math: Test the speed of the four basic integer math operations.


The tests are done on a ROM 03 GS with a TransWarp running at 13.75 MHz.
Please note that the figures below 13.75 MHz don't mean that the NC is
slowing down the TransWarp. This is the real effective speed compared
to a 1 MHz GS (Yes, I actually ran the tests at 1 MHz :) I get an average
speed of 11.1 MHz on these tests wo/ NC, and only Primes (13.57 MHz) and
Fibonacci (9.6 MHz) are slightly faster.

albert