aburto@marlin.UUCP (Alfred A. Aburto) (11/15/86)
0123456789012345678901234567890123456789012345678901234567890123456789012345 **************************************************** * Savage Benchmark Results * * 13 NOV 1986 * * Al Aburto/Lew Wolfgang/Larry Phillips/And Others * **************************************************** System CPU / FPP CLOCK Language Time Error (MHz) (sec) Abs(a-2500) Turbo-Amiga (68020/68881) 14.32 Absoft F77 V2.2B 0.4 2.0 E-12 HP 9000/320 (68020/68881) Fortran 77 0.7 3.2 E-09 HP 9000/320 (68020/68881) Pascal 0.7 2.8 E-07 Amiga (68020/68881) 7.16 Absoft F77 V2.2B 0.8 2.0 E-12 VAX-8600 Fortran 77 0.9 1.8 E-08 HP 9000/320 (68020/68881) C 1.0 2.5 E-08 VAX-11/750 Fortran 77 1.9 6.6 E-10 Masscomp (68010/ FPP) 2.1 3.2 E-07 DMS ( 8086/ 8087) Turbo-Pascal 3.8 1.1 E-09 Turbo-Amiga (68020/68881) 14.32 MetaComCo ABasiC V1.0 4.3 2.3 E+01 IBM PC-AT (80286/80287) ProFor F77 4.9 8.7 E-11 IBM PC-AT (80286/80287) Microsoft Fortran 77 7.2 1.2 E-09 IBM PC-AT (80286/80287) Turbo-Pascal 7.4 1.2 E-09 Amiga (68020/68881) 7.16 MetaComCo ABasiC V1.0 8.6 2.3 E+01 Amiga (68020/-----) 7.16 MetaComCo ABasiC V1.0 36.0 2.7 E+02 Amiga (68000/-----) 7.16 MetaComCo ABasiC V1.0 39.7 2.7 E+02 HP 9826 (68000/-----) 8.00 HP Basic V2.0 44.5 3.2 E-07 Amiga (68020/-----) 7.16 Absoft F77 V2.2B 59.7 2.8 E-07 HP Integral (68000/-----) Basic Interpreter 60.9 3.2 E-07 HP Integral (68000/-----) C 63.0 3.2 E-07 Amiga (68020/-----) 7.16 MS AmigaBASIC V1.0 67.0 3.2 E-07 Turbo-Amiga (68020/-----) 14.32 Lattice C V3.03 69.4 3.2 E-07 Amiga (68000/-----) 7.16 MS AmigaBASIC V1.0 73.0 3.2 E-07 Amiga (68000/-----) 7.16 Absoft F77 V2.2B 77.2 1.8 E-07 HP Integral (68000/-----) Absoft F77 100.0 1.8 E-07 Amiga (68020/-----) 7.16 Lattice C V3.03 139.0 3.2 E-07 Macintosh (68000/-----) MAC C 221.0 (?) Amiga (68000/-----) 7.16 Lattice C V3.03 234.0 3.2 E-07 Macintosh (68000/-----) DeSmet C 244.0 (?) Commodore 128( 6502/-----) 2.0 Basic Interpreter 256.0 9.0 E-04 Macintosh (68000/-----) Manx Aztec C 353.0 (?) *************************************************************************** Notes: (1) The Savage Benchmark, by Bill Savage, first appeared in Dr. Dobbs Journal, Sep 1983, Page 120. (2) The Macintosh results are from Byte, The Small Systems Journal, August 1986, Page 254. There appears to be a 'typo' in the published accuracy results. Exact result should be 2500.0 . (3) The Savage Benchmark requires use of IEEE double-precision variables and functions to obtain a reasonably small error. All the above were conducted with double-precision except MetaComCo's ABasiC where the variables were double-precision and the functions were computed only to single-precision. (4) The ABasiC interpreter with 68020/68881 is not a standard MetaComCo product. I modified the original ABasiC by MetaComCo to work with the 68020/68881 for my own purposes. I included these single precision results because I expect the double precision timings to be very close to these when using the 68881. (5) A number of people have requested that other systems be included (Sun 3/160, Zenith Z-248, Compaq 386 systems, Macintosh Fortran) but these are not readily available to me at this time. So if someone could contribute results for these systems or others (IBM PC-RT) it would be greatly appreciated. (6) These results are not official. Corrections, updates and additions are welcome. There are a number of interesting features that came out of these tests such as the poor performance of the Amiga and MAC C compilers with double-precision. (7) I also have results for a 512 DFT benchmark. These results show timings for the Amiga's Fast Floating Point (FFP) library which are indeed very fast if one doesn't need more than about 6 digits of precision (to start with). ****************************************************************************** Al Aburto
aburto@marlin.UUCP (Alfred A. Aburto) (11/15/86)
Here is the Savage Benchmark Program:
**************************************
* Fortran 77 *
**************************************
Program Savage
implicit double precision (a-h,o-z)
write(*,1000)
a = 1.0
iloop = 2499
do 100 i=1,iloop
a = dtan(datan(dexp(dlog(dsqrt(a*a))))) + 1.0
100 continue
write(*,1010)
write(*,1020) a
1000 format(5x,'Start')
1010 format(5x,'Stop ')
1020 format(5x,'a = ',f22.15)
stop
end
--------------------------------------------------------
Al Aburto
ilooali@navajo.STANFORD.EDU (Ali Ozer) (11/18/86)
In article <838@marlin.UUCP> aburto@marlin.UUCP (Alfred A. Aburto) writes: >**************************************************** >* Savage Benchmark Results * >* Al Aburto/Lew Wolfgang/Larry Phillips/And Others * >**************************************************** >System CPU / FPP CLOCK Language Time Error > (MHz) (sec) Abs(a-2500) >Turbo-Amiga (68020/68881) 14.32 Absoft F77 V2.2B 0.4 2.0 E-12 >VAX-8600 Fortran 77 0.9 1.8 E-08 >VAX-11/750 Fortran 77 1.9 6.6 E-10 >Amiga (68000/-----) 7.16 MetaComCo ABasiC V1.0 39.7 2.7 E+02 >Macintosh (68000/-----) MAC C 221.0 (?) >Commodore 128( 6502/-----) 2.0 Basic Interpreter 256.0 9.0 E-04 > ... Thanks for the benchmarks! I'm happy to see the Amiga up there. I ran the savage program on a Dec 2060 Mainframe, and it gave an answer to within 2.0 E-12 (as good as the best in the above table) in 1.6 CPU seconds (4 times slower than the Turbo Amiga!). One question concerning the benchmarks on the Amiga --- Are they cpu time or just plain real time? Tom Rokicki once wrote to the net that his 68000 version of Life was running 17% slower than he figured, and one reason people came up with was the multitasking overhead of the Amiga. For example, the Dec20 figure above is the CPU time --- The real time was more on the order of 6 seconds. Considering Amiga is a multitasking machine, the benchmarks should give the CPU time, if possible. In any case, I also ran the program using Basic on a Tandy PC-5 Pocket Computer. It gave the result to within 2.7 E-3 (almost as good as a C128) in 961 seconds (16 minutes!). Oh well, and that thing cost me 1/12th as much as an Amiga! Ali Ozer, ali@score.stanford.edu
aburto@marlin.UUCP (Alfred A. Aburto) (11/19/86)
In article <1094@navajo.STANFORD.EDU> ali@navajo.UUCP (Ali Ozer) writes: > >One question concerning the benchmarks on the Amiga --- Are they cpu time >or just plain real time? Tom Rokicki once wrote to the net that his >68000 version of Life was running 17% slower than he figured, and one >reason people came up with was the multitasking overhead of the Amiga. >For example, the Dec20 figure above is the CPU time --- The real time was >more on the order of 6 seconds. Considering Amiga is a multitasking >machine, the benchmarks should give the CPU time, if possible. The Amiga Savage results are not CPU time. But they were run with a no additional user generated tasks running in the background though. I thought this was a reasonable approach as those tasks that do interrupt the program were (I hope) necessary system tasks. Anyway you're right that the Amiga results are not the CPU seconds for the program run. I found that these other system tasks lower the Amiga's 7.16 MHz clock rate to an effective system clock rate of approximately 6.2 MHz which is near to what Tom Rokicki found (17% slower). l burto