Conference hydra::amiga_v1

600.0. "Savage Benchmarks" by LEDS::ACCIARDI () Fri Jul 17 1987 22:41

    The latest issue of Amazing Computing has an interesting review
    of the Turbo Amiga, from CSA of San Diego.  
    
    Some people have queried about the viability of an Amiga as a low-cost
    number cruncher, and these benchmarks are very interesting.
    
    The test used to measure number crunching power of the various system
    tested was the standard 'Savage' benchmark, which computes the value
    of: 
    
    	a=tan(atn(exp(log(sqr(a*a)))))+1.0
    
    for a loop of a=1 to 2499, and posts the time taken for the results.
    This expression is very masochistic, since the trig and log functions
    are computed from infinite series (remember Calculus III)?
    
    I've summarized some of the more interesting results below.
    Unfortunately, no benchmarks were taken for the MAC ][, which would
    seem to be the most comparible engine to the Turbo Amiga.  Notice
    also that the stock Amiga posted some very respectable times using
    AmigaBASIC!  The Atari ST did very well under the Absoft Fortran 77
    compiler.  Many C compilers for both the Amiga and the Atari ST
    were actually slower than BASIC!
    
    The tests results are listed in decreasing order of performance.
    
    System	CPU/FPU		Clock	Language   	Time 	Error
    				(MHz)			(secs)  (a-2500)
    ----------------------------------------------------------------------------
    VAX-8600	-----		-----	Vax Fortran     0.28    6.6E-10  
    Turbo-Amiga 68020/68881     14.32   Absoft F77      0.39    2.7E-12
    VAX-8650    -----           -----   VAX Fortran     0.40    6.6E-10
    Amiga       68020/68881      7.16   Absoft F77      0.78    2.0E-12
    VAX-11/780  -----           -----   VAX Fortran     1.16    6.6E-10
    IBM PC-AT   80286/80287      6.00   ProFor F77      4.90    8.7E-11
    Turbo-Amiga 68020/68881     14.32   AmigaBASIC      6.80    1.2E-9
    IBM PC       8088/8087       4.77   Microsoft C     8.00    1.2E-9
    Sun-3/160   68020/-----     16.67   Sun 3.0 F77     21.5    3.1E-7
    Amiga       68000/-----      7.16   TrueBasic(Comp) 65.2    3.0E-3
    Atari ST    68000/-----      8.00   Absoft F77      67.6    1.7E-7
    Amiga       68000/-----      7.16   AmigaBASIC      73.0    3.2E-7
    Amiga       68000/-----      7.16   Absoft F77      77.2    1.8E-7
    Atari ST    68000/32081      8.00   Megamax C      119.0    2.2E-8
    Amiga       68000/----       7.16   Aztec C V3.30  120.3    3.2E-7
    Macintosh   68000/-----      7.83   MAC C          221.0    -----
    C-128        8502            2.00   Basic Interp.  256.0    9.0E-4
    Macintosh   68000/-----      8.00   Megamax C      495.0    8.5E-7
    IBM-PC-XT    8088/------     4.77   Basica V2.10   895.0    3.0E-8
    
    The usual disclaimers apply, blah blah.  Results were copied here
    without any permission whatsoever.
    
    It has been said many times that these standard benchmarks test the
    compiler efficiency and not too much else, so don't postpone buying
    that VAX 8650 you've always wanted just because the CSA Amiga was a tad
    faster in this test. 
    
    I find it interesting that AmigaBASIC fared well against Absoft Fortran
    on the stock Amiga. Also, what on earth is a 32081 FPU chip used in the
    Atari/Megamax run? 
    
    These tests convince me of several things... an Amiga 2000 with the CSA
    Turbo card (already demonstrated) should be a real killer for around
    $3500.  Even the stock 7.16 MHz Amiga with the CSA Turbo card packs
    a whallop.  If only some good engineering software (ANSYS, Spice, Matrix-X)
    would get ported to the Amiga... sigh....