#: 19052 S12/OS9/68000 (OSK)
    09-Oct-93  13:22:43
Sb: #Review: 340 Upgrade
Fm: Zack Sessions 71532,1555
To:  all

I am the proud owner of an MM/1A, that is an MM/1 which has been upgraded
with the 68340 Accelerator Upgrade. This article is a description of the
upgrade, what is involved in installing it, what it is supposed to give
you, some gotchas I ran into installing it on my system, and some other
observations.

What it is:

        A "daughter board" which has the 68340 and a few support chips on it.

        Two replacement ROMs.

        One replacement PAL.

        Floppy disk with new kernel and drivers.

        Cost: $325

Optional:

        If you are one of the unfortunate ones whose IO board was equipped with
        74HC257s (there are three of the in a row on the IO board) instead of
        74AC257s, then you will need to replace the HCs with three ACs. (Ouch!
        virtually all IO boards which have the HCs, there are soldered in!
        I watched Kevin Pease replace these three chips in Bill Wittman's MM/1
        in less than 20 minutes, so it is not an impossible task!)

        Note: If your IO board does have the 74HC257s, you can stil upgrade
        without replacing them. You can use a PAL which supports a four cycle
        memory access (instead of the three cycle) and put up with a 20% loss
        in speed gains.

Installation:

        MAKE A NEW BOOT DISK FIRST!!! Instructions on how to do this is in
        the install docs.

        Remove the IO board. Replace the PAL on it. Replace the 74HC257s if
        necessary. Rejumper the IO board for "no wait states". Replace the
        two ROMs on the mother board. Remove the 68070 and treasure it
        forever. Set the speed DIP switch on the daughter board accordingly.
        Plug the daughter board into the empty '070 socket. Reinstall the
        IO board. Boot up from the new boot floppy. Customize your previous
        bootlist to use the new modules where needed.

The entire process (if you do not have to replace the 74HC257s) should
take no more than a half an hour to an hour, depending on how long it takes
you to recustomize your previous bootlist.

What is the 340 upgrade supposed to give you?

        More processing power. The 68070 was a backwards engineered 68000
        and therefore had numerous situations where the microcode was less
        efficient than an original 68000. The 68340 is a genuine Motorola
        product and is an improvment over the 68000 in microcode efficency.
        Most instructions take fewer (usually half as many) clock cycles to
        complete plus the 68340 has some of the 68020 instructions. Even though
        the "normal" clock speed of the 68340 will be 16.59 Mhz for most MM/1
        users, it still rates about twice as fast as a 15Mhz 68070. According
        to the UG docs, the 68340's clock is adjustable from 11.98 Mhz
        to 25.80 Mhz (although one user has reported he was able to
        set his down to 8 Mhz with a special system state program.)
        If you have slow DRAMS (ie, 100ns) you will only be able to
        reliably run at about 12 Mhz. 80Ns DRAMS will allow a clock
        speed of about 16 Mhz. The docs say that 60ns DRAMS should allow a
        clock speed of 20 Mhz, but Kevin Pease told me that all things
        considered (other chips on the boards, etc.) that one really
        wouldn't be able to get reliable operation above the 16.59 Mhz
        value.

        Better system IO since the DMA transfers are not limited to a size
        of 64K. The new scsi driver which comes with the UG apparently
        takes this new feature into consideration.

        More and better serial ports. The 68340 has three serial ports, being
        used as /t0, /t1 and /t5. These are improvement over the '070s two
        serial ports in that they look more like 68681 ports, they are full
        ports with hardware handshake, but CD is not currently implemented. A
        standard /t3-/t4 paddle is required to use /t5. (The header for it is
        on the bottom side of the '340 daughter board.)

        The new ROMs support booting your MM/1A from your hard disk.

        The new serial drivers now will time out on close, so that a modem
        port that is stopped by xoff can still be closed.

All in all, these advantages add up to a MM/1 which is roughly twice as
fast as it was before the upgrade. All of the "benchmarks" I tested
held true to this.

Gotchas which hit me were:

        There is an elusive bug which Carl and Kevin are both aware of which
        causes you to potentially have problems making floppy boot disks. If
        your floppy device is not inized (or has some path in the system open
        to it), then when os9gen tries to rename the file TempBoot to OS9Boot,
        it fails. I forgot the exact error message. There are two workarounds.
        One is to simply rename the file yourself and do a -q option on another
        os9gen command. (I found that this workaround did not work,
        intermittantly.) The other is to simply put an iniz /d0 command in
        your startup.

        Why do I even need to make a floppy boot disk, you may ask? Well,
        problem two is that apparently the boot code in the new ROMs ONLY
        perform "multi-sector reads". My hard drives, Maxtor 7213S, apparently
        do not support multi-sector reads, as I had to turn this bit off
        even with the '070 to get them to work. The result is when I try to
        boot from my hard drive, the boot code dies with a sector read error.
        Kevin has assured me that he will be able to fix this, but will need
        to borrow one of my hard drives to test the code with.

        My third and last major problem was that I could not format my hard
        drives. Since I have two drive exacly alike, I decided some good tests
        would be copying files back and forth, etc. So the first thing I did
        was to reformat /h1. (Of course I had done a backup first!! )
        But when format wanted to write LSN0, it would die, again, I forgot the
        exact error message. The solution here was easy. The original OSK
        disks from IMS included the format command from V2.3 and said that
        it was needed under certain circumstances, formatting floppies or
        something like that. So, my format command in my CMDS directory was
        the V2.3 version. The V2.4 version was in ROM, so whenever I ran
        format, it was coming from ROM anyway. Well the replacement ROMs which
        come with the 340 UG do NOT have ANY of those additional modules and
        program in it which were in the original ROMs. So, when I replaced
        my ROMs for the upgrade, I was now running the V2.3 version of format.
        Scrounging up my original floppies and getting the V2.4 version of
        format solved this problem.

Other observations:

There is a mouse driver for Microsoft mice for the /t2 port, but there
is not one yet for /t0.

There is NO sound driver AT ALL, yet. Supposedly there will be one soon,
one which takes the larger DMA transfer size into account.

The DIP for adjusting the the clock speed is on the bottom of the board,
and cannot be changed without actually removing the daughter board. But
one can write a system state program and adjust the clock rate on the
fly, the address and formula is provided in the docs. Maybe soon, Brian
White will post his utility for adjusting the clock to Delphi or somewhere.

I use the new /t0 for my modem port and it seems to function just fine. I
haven't tested /t5 yet.

Ulti-MuseE OSK seems to work just fine.

dEd (OSK) will not run without a Stack Overflow. One can workaround
this problem by increasing the stack space given to dEd on the Shell
command line, ie:

$ ded #100 filename

Autofollow mouse seems MUCH, MUCH smoother and responsive.

One can take advantage of the included 68020 instructions by one of the
following methods:

        Use GNU-C with appropriate switches
        Use Ultra-C with the appropriate switches

Using the 68020 instructions will make the MM/1A faster still. I would be
interested in seeing some benchmarks with this.

Remember that the compiled code will only run on a 68340, or 680x0 where
x >= 2. (Or possibly some other Motorola processor I am unaware of.)


Oh yeah, one last thing. The UG comes with a new shell. It doesn't seem to
be very different from the old shell, but I have noticed two differences.
One is that is used cd and cx instead of chd and chx, in fact if you try
to run a chd or chx command you'll get a 216 error! The other difference
is that when you fork a process at the shell level with a & on the end of
the line, the PID of the child process which gets displayed does not have
the plus sign (+) in front of it. I would have really liked it if there
had been SOME documentation on this new shell.

Conclusion:

I feel that the investment was well spent. I am very pleased with the speed
increase in my MM/1 and can live with the minor problems which have not
been solved yet.

------------------------------------
 Zack C Sessions
 ColorSystems

 via InfoXpress/OSK by Bill Dickhaus

There are 2 Replies.