SPEED COMPARISONS
-----------------

If SysInfo does  not have a test that  you would like to see,  let me know
and I will do my best to add it in for you.

THE SPEED  COMPARISON CODE HAS BEEN  WRITTEN TO GIVE A  FAIR INDICATION OF
THE SPEED  OF A PARTICULAR  AMIGA. IT USES  A COMBINATION OF  ALL MOTOROLA
INSTRUCTIONS TO  BOTH REGISTERS  AND MEMORY IN  A PERCENTAGE  THAT POPULAR
AMIGA PROGRAMS AND COMPILERS HAVE USED.

The A500  - A600  STD comparison  is against  a PAL  A500 or  A600 totally
unexpanded or expanded to 1 Meg chip only (ie. no Fast Ram).

The B2000  EXTRA RAM  comparison is  against a  PAL B2000  Rev 4.4  with a
Microbotics 8-UP Fast Ram board fitted with 80ns DRAMS.

The A1200  68EC020 comparison is against  a standard PAL A1200  as shipped
from Commodore with cache enabled. The  reason for the difference in speed
to the A2500,  is because the A1200  is shipped with CHIP  ram only. There
will be a significant speed increase when fast memory is added, especially
32 bit memory such as the Microbotics MBX 1200 board.

The A2620 comparison is against a standard A2500 with a A2620 card running
at 14.3 MHZ. All caches and bursts on.

The A3000 25 MHZ comparison is against  a standard PAL A3000/25 MHZ with 4
Megs of  1MX4 SCRAMS  80ns 32  bit wide, and  2 Megs  of CHIP.  Caches and
Bursts on  except Data Burst (default  under V2.04 V37+). Ramsey  mode was
set at default, BURST ON and STATIC COLUMN OFF.

The A4000/040  25MHZ is  against a  standard PAL  A4000 with  the standard
configuration of 2 Meg Chip and 4  Meg Fast memory. All cache modes are at
boot default  under 68040.library 37.10  under AmigaDOS V3.00  - Kickstart
39.106, with setpatch 39.7 installed.

CPU MIPS (Million Instructions per  Second) This test calculation has been
coded from  all information  I have been  able to find  to date.  It seems
quite debatable how this is supposed  to be coded and I received different
information from people.  It does a very large loop  performing a total of
4.25  million  instructions  ranging from  general  instructions,  divide,
multiply, logic shifts,  rotations etc. I CANNOT AND DO  NOT GUARANTEE ITS
ACCURACY, except to say it is accurate  to the point, it did do the number
of instructions per second that it  stated. Comments on this approach or a
better one are most welcome.

FPU MFLOPS (Million Floating Operations  per Second) This test calculation
was similar  to the above one,  in that information on  the recognised way
was very sketchy. If  an Amiga does not have an FPU, N/A  will be shown in
this field  as any performance test  would be meaningless. The  final code
shows known boards to  be at or around thier advertised  speeds. It does a
very large  loop performing  instructions that take  an average  number of
clock cycles, times the loop then displays the result. I CANNOT AND DO NOT
GUARANTEE ITS ACCURACY. If anyone has a better routine for this I would be
very interested.  This routine is not  compatible with a 68881  on a 68000
system such as the Phoenix board and will show N/A.

Comments