🚀 go-pugleaf

Some additional remarks from a PAK (68020/88881) owner:

In article <8803280338.AA25219@ucbvax.Berkeley.EDU> BLANK@BRANDEIS.BITNET (David N. Blank * BLANK@BRANDEIS.bitnet) writes:
>Howdy,
>  This message is being forwarded for a friend that does not have net access
>at the present time.  Please do not respond to me, as I am an Amiga owner,
>and have negative quantities of interest in machines with 2 letter names.
>         Peace,
>           David N. Blank
>           BLANK@BRANDEIS.bitnet || DNB@BRANDEIS.csnet

Lots of stuff deleted

>The 68020 card is called the "PAK68", and they have interfaced it to various
>computers in addition to the ST. It has a 12mhz 68020 and 68881. The article
>for the PAK68 itself was in the 8/87 of c't; the ST interface article was
>11/87. The article this issue is only about the software hacks needed to get
>it running under GEM.

This ST interface is almost nothing. The original article describes a
68020 board which you can stick into an 68000 socket. (on many systems,
not only st, will even work for amiga's :-); some problems for other
systems are mentioned in the 4/88 issue).
What is called an ST interface here, is a small switching board, which 
can contain both 68000 and 020. You can select one of them by means of
a switch (some critical lines are switched by means of CMOS ic's; if I
recall correctly 43's)

Note that you can stick either 12.5 or 16
Mhz components on the board. However the 68020 still has to run on the
8 Mhz provided by the ST. You do NOT get a 12 Mhz 68020, unless you apply
some additional hardware hacks! However, it is possible to use the
68881 asynchronous to the 020, thus having it run at 16 Mhz.
The board does allow different clocks to be fed to both chips, and has
room for one oscillator.
>
>There were 3 basic problems: (1) the MOVE SR, <ea> instruction, which is
>privileged on the 68010 and 68020, (2) GEM is dependent on the exception stack
>format of the 68000, and (3) the F-Line routines of the AES use self-modifying

Actually line F has another problem. The 020 thinks line F codes are for
the 68881, and does not call the appropriate handler. This problem is
much more a pain in the ass.

>code. They present both source code for fixups and ROM patches for the 2/86
>and "blitter" ROMS. They report that 100% of TOS programs work! They present

Note here that the ROMs are German roms (different keyboard layout and
messages). Better start from the source code. (anyone having the source
of a disassembler for me??? You need to find all line F calls.)
Additional note: they claim that almost 100% of the programs work.
Exceptions are programs that call line-F directly, or perhaps that are auto
booting (games).

>this benchmark for 100 iterations of the sieve (megamax C):
>
>        cpu        seconds      relative speed increase
>
>       68000        23.6             ----
>       68010        22.8             3.5%
>       68020        11.9            98.3%

The gain here must be due to the fact that the main loop fits in the
cache. This result is NOT typical. Expect percentages around 10 or 20.

>This version does not use the 32-bit memory add-on which they described in
>the 1/88 issue. They expect another 100% speed increase with the faster and
>wider RAM. They did not present a benchmark using the 68881. However, they

The add-on memory is SRAM ($$$). No add-on DRAM exist.

>did say that GFA Basic users with version 3.0 or higher could directly access
>the 68881. So I would _assume_ that the 68881 works, although I can't read
>enough of the article to see what's going on.

I do not know anything about the GFA basic stuff (in fact I do not know
anything about any basic). However, it shouldn't be difficult to create
your own fpp c library.

>The straight PAK68 board sells for 1098 dm (including the 68020 and 68881),
>but this doesn't include any 32 bit ram or the Atari-specific interface.

However, you can also sell the bare PCB from Heise Verlag, which is much
more interesting if you can obtain the 020/881 at some reasonable price.
Other hardware needed: about 11 TTL chips (AS flavour), some sockets,
jumpers etc.
The 68020/68881 board is 49 dm, the interface I believe 29.
Roms go for about 200 dm, rom images on disk for 30 dm.
Copies of the original articles do dm 5.
I don't have order numbers handy, but if you really want to order drop me
a note and I'll dig it up.
Note: in order to operate properly the 68881 is not required.
If you are on a low budget, you can go for the 68020 alone, and leave the
68881 for later.

Disclaimer: I have no interest in C'T or Heise Verlag. I'm only a
reasonably happy customer

Remainder of included message junked.

Final remark: Apart from the bootstrapping, the PAK board will run Minix
without any modification.

Hope this informs you all a little more.
-- 
Frans Meulenbroeks
	Centre for Software Technology
	...!mcvax!philmds!prle!cst!meulenbr   or   ...!uunet!prlb2!cst!meulenbr
        or perhaps   meulenbr@cst.prl.philips.nl