home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Amnesia 7
/
Amnesia - Issue 07 (1991-11-23)(Eclipse).adf
/
text
/
68020review
next >
Wrap
Text File
|
1989-05-09
|
12KB
|
242 lines
Here is a review of Solid State Leisure's A5000-16 68020 accelerator board,
which I got for my Amiga 500 a week ago. Before I start, I just want to mention
that I had to wait almost ten weeks for it to arrive, and telephoned them
(long-distance) several times during that period. What happened was that they
sent out one board by registered post, which never got to me. Since it was
insured, they sent me out a second one, as they would be able to claim
compensation for the first one. The second one never got to me either. In actual
fact, both were eventually sent back to SSL by Irish Customs (I live in Ireland;
SSL are in England, which is the country next-door to us; look at a map...), who
for some unknown reason, refused to send it on to me. I can only assume that SSL
made a mess of the Customs declaration docket on the parcel, as I have received
a lot of other goods from England in the past, including a hard drive, with no
problems at all. Anyway, they finally sent the third board out by courier, which
did get through!
Hardware:
-------------
16.67Mhz 68020, with space for 4Mb of 32-bit RAM (comes with 1Mb as standard,
uses 256*4 80ns DRAMs, which is handy if you have some in your A590/A2091),
space for a 68881/2 maths coprocessor, but no MMU. All this sits on one small
board, which plugs into the 68000 socket in your A500/A2000. The original 68000
chip fits onto the accelerator board, and is used in the 68000 fallback mode.
You don't need a new PSU to use this board, even if it means you'll end up with
5Mb inside your Amiga, as the extra RAM chips take a minimal amount of power.
There is an option to copy your Kickstart ROM into fast 32-bit RAM, and the
board will automatically remap all the Amiga's ROM accesses into RAM, so
anything which relies on the OS routines (including windows) will run much
faster. This currently only works with 1.2/1.3 Kickstart ROMs, but an update to
the software supplied will support KS2.0, and will be available soon.
The memory is *not* autoconfig; you must run an AllocMem program in your
Startup-Sequence, which searches for 32-bit memory, and adds any found to the
system's memory list.
Installation:
-------------
Probably not for those who don't feel confident about ripping chips out of their
Amiga (I got a friend to do it for me). The first time we switched on with the
'020 installed, the screen went a worrying shade of green, but this was probably
because the board wasn't properly seated in the socket. (As an aside, ever
wanted to know what happens when you turn your Amiga on with no 680x0 chip at
all? - Absolutely nothing! (You see a grey screen.) 8-) )
I was able to use some of the RAM chips from my A590, and these go into the
4 1Mb banks on the board, so you can have from 1-4Mb of 32-bit RAM. (I have a
total of 3Mb of 32-bit RAM, 1Mb of 16-bit fast RAM, and 1Mb of chip RAM, giving
5Mb altogether.)
Speed:
------
SSL claimed a 500% (5 times) speed increase, and true enough, most benchmarks
will give this. Mips gives a result of 4.969519 (as opposed to 0.847750) which
is almost six times faster. AmigaBench gives a result of 3561 Dhrystones/sec
(as opposed to 112), but the optimised 68020 version gives 5694. The various
benchmark programs included with the board (CalcPi, Ronin CPU speed, Whetstone,
another version of Mips) give 4-5x speed increases.
However, benchmarks like these are not what it's all about, so I did some
"real-world" tests, as these are the ones you will find useful/interesting:
68000 68020 Speed increase
Time to LhArc DPaint to RAM: 3m00s 1m06s 2.7
Time to un-LhArc it 1m09s 0m22s 3.1
Time to PowerPack a 200K file 1m53s 0m34s 3.3
Time to 'echo' a 400K sound sample in AM3 2m43s 0m50s 3.2
Time to decompress a disk using DMS 4m20s 1m26s 3.0
Time for HamSharp to convert a pic GIF->IFF 1m35s 0m30s 3.1
Time for TurboGif to display a GIF pic 11.2s 4m.3s 2.6
Time to open a dozen windows on Workbench 26.7s 18.0s 1.4
Time to close them all 29.1s 14.0s 2.0
Time to draw a Mandelbrot set 26.6s 5.9s 4.5
Oi! Where's my DMA gone??
-------------------------
There is one serious problem with the board: you can't DMA from your hard drive
into its 32-bit memory. There has been some discussion of this problem on
Usenet, and apparently while the FileSystem manages to get around it, it doesn't
do it very well. Just to show what I'm talking about, here are the results from
DiskPerf2, running first in normal 68000 mode (DMA-ing into ordinary 16-bit fast
RAM), then in 68020 mode (failing to DMA into 32-bit RAM, so it reads 512 bytes
at a time into 16-bit RAM then CPU copies them up to 32-bit RAM). My hard drive
is a Quantum LP52S - a very fast SCSI HD - in an A590.
DiskPerf2. Testing Files:
Create Files: 23 files/sec. Directory Scan: 117 entries/sec.
Delete Files: 62 files/sec. Seek/Read Test: 94 seek/sec.
Read/Write Speed Test: (bytes/sec.)
Buffer: 512 Read: 30,169 Write: 22,036
Buffer: 4k Read: 221,405 Write: 169,892
Buffer: 8k Read: 325,644 Write: 255,252
Buffer: 32k Read: 382,691 Write: 445,823
Buffer: 64k Read: 476,625 Write: 562,540
Buffer: 256k Read: 576,140 Write: 695,342
(Note: This was on a slightly fragmented partition; I sometimes obtain speeds
of up to 800-850K/sec.)
DiskPerf2. Testing Files:
Create Files: 15 files/sec. Directory Scan: 125 entries/sec.
Delete Files: 61 files/sec. Seek/Read Test: 80 seek/sec.
Read/Write Speed Test: (bytes/sec.)
Buffer: 512 Read: 26,211 Write: 19,772
Buffer: 4k Read: 39,954 Write: 29,049
Buffer: 8k Read: 40,398 Write: 29,293
Buffer: 32k Read: 40,249 Write: 27,643
Buffer: 64k Read: 39,991 Write: 27,902
Buffer: 256k Read: 40,077 Write: 28,352
As can be seen, this problem occurs when manipulating large files; you won't
notice any difference when loading/saving small files, such as small utilities,
text files etc. But for large files, you're talking about a 3-6x
speed _decrease_. I copied out 600K of data (mostly one big file) from my hard
drive to RAM: before installing the 68020, and it took 6.9s. With the 68020,
it took 21.7s. Copying from one partition to the other was twice as bad: 8.5s
for the 68000, 50s (!) for the 68020. Reading 2Mb sound samples in AudioMaster
is now a joke. For some operations, I have timed speeds of as low as twice that
of a floppy, cough, choke...
This problem is something which you can live with, as of course even the worst
cases are faster than floppy, and it won't affect you very much unless you like
messing around with large files, or frequently boot very large applications
(such as ProPage). But it is something to bear in mind if you are considering
purchasing this board. Before I ordered it, I asked SSL about the DMA problem,
and the man I was talking to told me it wouldn't really affect me as long as I
left some ordinary 16-bit fast RAM in the A590 (which you can DMA into) as a
buffer for the FileSystem to use. Hmm... I think he was wrong!
Just as an aside: I ran DiskPerf on the RAM disk, and was suitably impressed
by the results. First 68000, then 68020 with its 32-bit RAM...
DiskPerf2. Testing Ram Disk:
Create Files: 14 files/sec. Directory Scan: 15 entries/sec.
Delete Files: 28 files/sec. Seek/Read Test: 360 seek/sec.
Read/Write Speed Test: (bytes/sec.)
Buffer: 512 Read: 212,606 Write: 174,066
Buffer: 4k Read: 840,205 Write: 548,418
Buffer: 8k Read: 967,321 Write: 599,871
Buffer: 32k Read: 1,092,266 Write: 647,269
Buffer: 64k Read: 1,106,092 Write: 658,653
Buffer: 256k Read: 1,125,081 Write: 668,734
DiskPerf2. Testing Ram Disk:
Create Files: 39 files/sec. Directory Scan: 47 entries/sec.
Delete Files: 80 files/sec. Seek/Read Test: 370 seek/sec.
Read/Write Speed Test: (bytes/sec.)
Buffer: 512 Read: 255,750 Write: 224,438
Buffer: 4k Read: 1,899,594 Write: 916,587
Buffer: 8k Read: 2,702,515 Write: 1,078,781
Buffer: 32k Read: 3,912,597 Write: 1,248,304
Buffer: 64k Read: 3,912,597 Write: 1,337,469
Buffer: 256k Read: 4,599,017 Write: 1,351,257
Compatibility:
--------------
Very good, actually. Almost all productivity software runs perfectly under the
'020. There seems to be only one problem with the board with respect to
compatibility: while it doesn't have an MMU, some software thinks that it does,
and promptly crashes when trying to find out more about it.
A-MaxII crashes because of this, as does Nic Wilson's SysInfo. Both crash with
a "Coprocessor Protocol Violation" error (GURU 8000000D). Another program once
told me I had a 68020 with a 68851 MMU, but has crashed every other time I tried
to run it. This is quite worrying...
Another program which doesn't like the board is HD-Toolbox. When you boot it,
you are greeted by a requester telling you that some drives have been
added/removed from the system, and to click on "Save changes to drive" to tell
other drives about this. But on clicking the "Partition Drive" button first,
you see that all your partition data has been lost, and instead has been
reconfigured to the default 2 partitions (25Mb each in my case). Click on
"Save changes", and 0.5 seconds later, you are left with one very empty hard
drive!! You don't even get an "Are you sure?" requester! Since the program works
with the A3000, I can only assume there is something wrong with the 68020 board
to upset HD-Toolbox in this way. (It works in 68000 fallback mode, though.)
IntuiTracker and EdPlayer don't like faster processors, and screw up the music
slightly, but ModuleMaster works fine, as it uses the new ProTracker
play-routines. Apart from that, everything else seems to work fine.
I can't say much about games, as I don't exactly have terribly many of them.
Two I did try were R-Type2 and F-18 Interceptor. The former seems to work fine
with the added speed (it crashes on the A3000, though), and the latter *blazes*
along if you get it into 32-bit RAM. If you've ever seen it on the A3000, you'll
know what I'm talking about!
Many demos don't work on the '020. I tried a total of nine. The ones which
worked perfectly were Coma, Dreamscape, Mesmerized, Neutron Dance and
Phantasmagoria (although the scrolly message at the start of this was a bit
upset). The ones which didn't were Angels (starts out OK, then there are slight
glitches, then finally it crashes), Mental Hangover (crashes after a while),
Seven Sins (crashes fairly quickly) and Substance (lots of glitches, although
their fractal routine runs much faster...).
There is also a 68000 fallback mode. You operate this by clicking on an icon,
which reboots the machine using the 68000. Unfortunately, there doesn't seem to
be any way of getting out of this mode, other than turning the machine off then
on again! And this doesn't suit me at all, especially with the hard drive
attached: once I left it off for 15 seconds, turned it on, and was greeted by a
yellow screen, which wouldn't go away. (How long *should* you wait after
powering-down when you have a lot of memory installed, and an expensive hard
drive??) Also, the board's memory cannot be accessed in fallback mode, as it is
outside the 16Mb address space...
That's the end of this review! If you want to know any more about the board, or
if you want SSL's address (they also do 68030 boards - 25Mhz and 40Mhz), feel
free to e-mail me.
Barry.
bmccnnll@vax1.tcd.i