Conference hydra::amiga_v1

4412.0. "A review of Adspeed Accelerator" by SHARE::DOYLE () Tue Jan 08 1991 09:30

The ICD AdSpeed CPU Accelerator for the Amiga 500

Earlier this month, ICD provided us with a pre-release version of the
AdSpeed Accelerator so Safe Harbor could test it and announce the
results.  I call this unit a pre-release, even though it appears to be a
production unit, because we are still waiting for ICD to begin volume
shipments.  Despite this label, the board layout is conspicuously
missing any jumpers or modifications that seem so prevalent in early
versions of Amiga hardware products.  AdSpeed appears to be a very well
engineered high-tech produ ct that uses surface-mounted integrated
circuits.  These circuits are mounted between the legs of the 16 MHz
68000 that is provided by ICD.  I was surprised that ICD went to all of
this trouble but it guarantees that the AdSpeed can be installed in any
Amiga 500, regardless of the Amiga's circuit board layout.  During
installation of comparable products, I have encountered Amiga components
that interfered with installation but ICD has eliminated any chance of
interference by reduc ing the size of the AdSpeed to fit within the
confines of the 68000 socket.

Installation of AdSpeed was as simple as opening the 500, removing the
metallic RF-shield and replacing the existing 68000 with AdSpeed.
AdSpeed's compact assembly appears to be more rugged than a plain
integrated circuit so installation should be easy for anyone that has
had prior experience installing integrated circuits.  As always, opening
your Amiga voids CBM's warranty so do not attempt to install AdSpeed
unless you are sure you know what you are doing.  As supplied, AdSpeed
was configured to power-u p in the accelerated mode (14 MHz) so initial
testing consisted of supplying power to the Amiga and verifying that
"workbench hand" appeared.

According to the more than adequate documentation supplied with AdSpeed,
AdSpeed can be jumpered for 7.14 or 14.28 MHz operation.  A toggle
switch can be connected to AdSpeed to provide a hardware toggle so that
speed can be adjusted at any time, not just at power-up.  AdSpeed also
comes with a software program that can toggle operation speed between
7.14 and 14.28 MHz regardless of the condition of the switch.  This
program runs from the Workbench or CLI and provides a display of the
current CPU speed.  Sp eed can be toggled with a mouse click.

One of AdSpeed's major advantages over competitive accelerators is
AdSpeed's use of a real 68000 CPU chip.  Programs that do not run with a
68020 or 68030 CPUs should operate properly with AdSpeed.  In the few
cases where a program has troubles with a faster CPU clock, AdSpeed can
be made to run at the standard 7.14 MHz clock speed insuring
compatibility.  AdSpeed achieves its acceleration by running its CPU at
14.28 MHz, twice the normal clock speed, and by using a 32K Byte
instruction cache.  Unlike previ ous attempts to accelerate the Amiga by
using faster CPU clocks, the cache makes all of the difference!  Since
programs often consist of repetitive operations, caching program
instructions in fast static RAM allows the faster clocked CPU to avoid
the bottle-necks associated with slowing down to 7.14MHz to access the
normal Amiga memories.  If the information needed by the CPU is in the
faster cache memory, operation can be twice as fast as a standard Amiga.
Caching is one t echnique used to speed up operati ons in the 68020,
68030 and 68040 CPUs.  AdSpeed has brought caching to the 68000.

To test the relative speed increase associated with AdSpeed, I used
Lamonte Koop's Amiga Intuition Benchmarking program to measure our Amiga
500's performance with and without AdSpeed.  While AIB is not perfect,
it appears to have some problems relating Amiga 500s to the Amiga 2000
reference, it is suitable for the task at hand.  In addition to AIB, I
decided to measure the time required to unlharc (extract) AIB from its
archive form on the Supra harddrive to the RAM: drive.  I also measured
the time it too k LHARC to compress the entire DH0:c directory to RAM:.
Extraction and compression are two operations that are quite repetitive
in nature and which should show improvement with a cache.  I also use
CSA's bench mark program to measure integer and floating point operation
as a check against Lamonte's program.


Our test system consisted of an Amiga 500 with KickStart 1.2, a 501 type
512K memory expansion and a Supra 500XP harddrive system with 1.5 MBytes
of FAST RAM.  As configured, this system posted the following
benchmarks:

                           Without AdSpeed        With AdSpeed        Relative
                           ---------------        ------------        --------
AIB 2.01 (normal mode):

   Write Pixel:            33.86 seconds  81.2%   18.78 seconds 146.4%  1.80
   Sieve:                  50.06 seconds  95.9%   26.90 seconds 178.5%  1.86
   DhryStone:              623 Dhry/sec   93.5%   1131 Dhry/sec 169.8%  1.82
   Savage:                 99.16 seconds  88.7%   53.42 seconds 164.7%  1.86
   Sort:                   13.00 seconds  93.1%    6.72 seconds 180.1%  1.93
   Matrix:                 86.74 seconds  95.4%   43.04 seconds 192.3%  2.02

AIB 2.01 (selfish mode):

   Write Pixel:            30.14 seconds  91.2%   17.54 seconds 156.8%  1.72
   Sieve:                  46.60 seconds 103.0%   25.64 seconds 187.3%  1.82
   DhryStone:              668 Dhry/sec  100.3%   1181 Dhry/sec 177.3%  1.77
   Savage:                 91.86 seconds  95.8%   50.82 seconds 173.2%  1.81
   Sort:                   12.06 seconds 100.3%    6.36 seconds 190.3%  1.90
   Matrix:                 80.62 seconds 102.7%   40.92 seconds 202.2%  1.97

UnLharc DH0:AIBB201.LZH to RAM:
                           20 seconds             11 seconds            1.82

Lharc DH0:c to RAM:
                           15 minutes 34 seconds   9 minutes 34 seconds 1.63

CSA Benchmarks:
        Integer:           1.0                     2.0
        Floating Point     1.0                     1.9

As indicated by these tests, AdSpeed can significantly boost the
performance of a standard Amiga 500 system and is definitely the product
of choice if you want maximum compatibility with existing software.  As
with all benchmarks, keep in mind that your mileage may vary depending
on the applications you use.

Dennis Hoffman - SYSOP, Safe Harbor BBS

12/31/90

        It is the same unit (a guess on my part) that is available for
        the ST, where it works very well also.
        
        Bob

Does anybody know when/if ADSpeed will be available for the 2000?

	As far as I know about the board - and if you read the description
	carefully - I would say that this board fits into any 68000-based
	AMIGA, 'cause you only remove the original CPU from its socket and
	place the board into it. And if you have any worries about not enough
	space --> there is more room around the CPU in the A2000 than in the
	B-52.

	If anyone knows it better, feel free to correct my statements.

	
	-hg