Amiga Format CD 48/Amiga Format CD48 (1999-12-13)(Future Publishing)(GB)(Track 1 of 2)[!][issue 2000-01].iso/-in_the_mag-/workbench/gvpsim64/gvp_ram32bit.english

Text File | 1996-06-25 | 23KB | 605 lines

(Author's note: this was an article I originally wrote for the french
magazine AmigaNews, and printed in march 1996 issue; given my command
of English language, I did my *best* to translate it!)

[Uploader's note: I reformatted this file and made a very few minor
modifications to it to make it easier to read in an 80-column window.
Carl Mueller, 6/7/96]


<<<<<<<<<< GVP Combo boards "G-Force" '030 (Combo 325/340/350) >>>>>>>>>>

Among the many accelerator boards available for our good old A2000s,
GVP boards have always ranked among the best, thanks to their
performance: high speed 32-bits DMA memory (Direct Memory Access for
highest possible data exchange rates, whose design "relieves" the
CPU's job), IDE or SCSI hard-disk controller (DMA too), extra parallel
port etc.

Most of the competitors' boards lack many of these functions; a
missing DMA, for example, forces the CPU to spend all its time
exchanging data by itself, making multitasking drastically slow down
during hard-disk transfers for example.

But no one's perfect: while G-Force '030 boards studied hereafter (also
called "Combo" for they mix CPU, FPU, 32-bits memory and SCSI
hard-disk controller altogether on a same board) are this powerful,
they require a specific kind of memory that costs nearly as much as
the board itself.

In the following article, I will do my best to show you clearly how to
make your GVP "love" a classic, hence much cheaper, memory for
PeeCees.


***** So-called "32 bits" memory

Nearly each accelerator board has its own memory, expandable (up to a
certain amount) or not, to which the board has a priority access, much
faster than any other CHIP or FAST memory. This extra fast memory is
called 32-bits because the CPUs run by such accelerator boards,
68020/030/040, all have a 32-bits wide data bus.

Another slowing factor, when accessing CHIP or FAST 16-bits memory, is
that this r/w accesses must be done thru the Zorro-II bus, while
accesses to the 32-bits memory of the board is straightforward with no
'bottleneck'.

As Ralph Babel explains in his Guru ROM manual, some boards, like the
GVP Combo itself, configure part or all of their 32-bits memory
within the Zorro-II address space (16MB) corresponding to the "basic"
68000, making this memory accessible thru DMA to Amiga's custom chips
(Agnus etc.) and other Zorro-II boards. High-speed DMA transfers are
driven by the on-board electronic logic, in either 16 or 32-bits wide
accesses, transparently for both the CPU (32 bits) and the "external"
devices thru the Zorro-II bus (16 bits).

As a conclusion, we can consider 32-bits memory as some sort of FAST
memory compared to CHIP memory: in fact, "FAST" memory has a
preferred, notably faster access to CPU 68000 than "CHIP" memory,
which is shared by nature between 68000 and Amiga's custom chips,
leading to slowdown. All FAST memory expansion board's owners will
tell you.

We can now see that it is our interest to have as much 32-bits memory
as possible on the accelerator board itself. An easy speed test would
be GvpMemTest (supplied with all GVP boards): 32-bits memory is read
and written at blasting speed compared to FAST or CHIP.

And the faster the memory accesses are, the faster all programs run!
It's rabbit versus tortoise.


***** What about bytes versus 32 bits ??

Don't let expressions like "1MB of 32-bits RAM" mislead you. How come
bytes (== 8 bits) and 32 bits can be mixed together ??

There *is* an historical reason to this: first microprocessors (CPU)
used to work on 8 bits wide data, then people designed memory in
blocks of 8 bits, called bytes. Today, most of the microprocessors
work on 32 bits wide data, but the old way of calling memory size
still remains.

We call a "word" the data size that each CPU is able to process at a
time: in general, from 8 to 64 bits. Then each word "weights", from
CPU's point of view, 1 or more bytes: for example, a CPU 68030 works
on 32-bits words, hence 4 bytes at a time. Therefore a Combo with 1MB
of 32-bits RAM means: 1MB of RAM (this is 1 mega-bytes, 1 mega x 8
bits) 32-bits wide, thus 1 mega x 8 / 32 = 256 kilo-words of 32
bits. "4MB of 32 bits RAM" means, the same way, "1 mega-words of 32
bits". Et cetera.

Be aware that, under Workbench, the Amiga only displays the total
amount of memory available in bytes, might it be 32 bits, CHIP or
FAST. Other softwares like Sysinfo, can tell one from another, and
will display the exact type of memory.


***** Special 32-bits memory for GVP Combo

To take full advantage of an accelerator board, the main point would
be to increase its 32-bits on-board memory. Some old memory boards,
like C= A2058 (Zorro-II), made use of "discrete" chips, to add by
blocks of 8 (x4) or 32 (x1) to always get these 32-bits wide data,
without which it would never work.

Other boards make use of SIMMs (Single In-Line Memory Module) 30 pins:
old models "short-sized" of 8 bits, by 4 (8x4=32 bits again and
again); or 72 pins, such as the new "long-sized" SIMMs of 32 bits (36
bits with parity) found in any recent PeeCee.

Some other boards even use specially designed memory, like SIMMs
Chipack 3240 of 4MB for GVP boards (not only for the Combo
models). And here comes the well-known money-making technique "made in
GVP": those Chipack 3240 cost actually big bucks. Ouch.

Moreover, they are special in several points:

  - their access time is 60ns, faster than most of the memory
    available at computer stores (usually 70ns or more); this speed is
    required to achieve the fastest possible memory accesses, with as less
    wait states as possible (which slow down the CPU); good job indeed
    GVP!!

  - their connector width is 64 pins, and their pinout is GVP-specific,
    making them totally incompatible with "common" SIMMs 72 pins;

  - the organization of the memory chips on the Chipack module
    itself is GVP-specific.

This means that the proud owner of a GVP board is *forced* to purchase
these Chipack 3240 to increase the board's memory, because no one but
GVP sells them. And no specific product sells for cheap: in France,
one of these 4MB modules might cost up to 1990F (around US $400). No
comment! Paying this price these days is suicidal: 16MB memory costs
less!! But since it is impossible to use non-GVP memory, there is no
alternative.

Impossible ?? Well, I'm going to prove you the contrary!


***** Needful things

WARNING!! The experiment takes quite long but isn't too difficult to
make. However, you must be familiar with the use of a soldering iron
and know how to make precise solderings, since the SIMM module itself
will need to be hacked. Money-wise, it's very worth it, but YOU are
responsible for any possible damage!

For hardware, you will need:

  - a SIMM module 4MB x32 (cf previously) 72 pins, speed 60ns as
    specified by GVP; such 60ns SIMMs are still not very common, be
    persistent!

  - a 72-pins SIMM socket; rather than soldering wires directly on the
    SIMM, the socket will come handy;

  - a low power soldering iron (30W) with a thin tip (1mm), and
    another one more powerful (40W or more) with a tip as thin as possible
    (2mm);

  - some wire, such as the "grey" one used to make IDE, SCSI, floppy
    cables.

  - some wire a bit thicker, of 2 different colours, to drain power
    signals VCC and VSS;

  - bits of wrapping wire (extra thin);

  - soldering, plyers, X-akto knife etc. and also a "beeper"
    (continuity tester; any multimeter, even the simplest possible, has
    this function).

TO AVOID LIKE THE PLAGUE:

  * SIMMs of 70ns or above (I did *not* want to dare the devil and
    hack a SIMM slower than 60ns, to realize in the end that it didn't
    work!);

  * old SIMM 30 pins, 8 or 9 bits wide (there are 256KB, 1MB and even
    4MB); 4 would be needed to get 32 bits wide memory;

  * SIMMs bigger or smaller than 4MB;

  * big soldering and soldering irons like a blowtorch.


***** About SIMM 4MB 32 bits:

Take care to purchase a SIMM 32 bits with NO parity, since it is
useless in our case; a SIMM 36 bits is seldom found and costs more, as
it is maily used by old PeeCees, like IBM PS/1.

Also, prefer a SIMM 32 bits but on which there still are the EMPTY
pads ready for the extra parity chips: we will take advantage of such
pads and tracks, left open, later. Ask for a friend's help: we
literally "see" that there is some empty place left on the module. It
is very commonly found and makes memory makers' life easy, to
manufacture 36 bits modules from 32 bits modules with the same PCB.

From now on, I will have to explain the full experiment in every
little detail, which might be tedious to follow but MUST be told,
should you make any mistake.


***** "Common" SIMM's ways of working

Any dynamic RAM (DRAM), found in every computer on earth, bears 4
command signals:

    /RAS (Row Address Strobe)

    /CAS (Column Address Strobe)

    /WE (Write Enable)

    /OE (Output Enable)

These signals are sent to the memory chip in a specific order (I won't
explain any deeper), and then the data bus either writes the data into
the chip, or reads the data from the chip. Addresses are multiplexed,
by rows and columns, like a 10 bits matrix A0...9.

On a 32 bits SIMM 4MB, there are 32 data bits split into 4 blocks of 8
bits each (4x8=32) hence 4 blocks of 1MB. Each 1MB block is made of 2
memory chips of 1Mbitx4. This makes 2 chips x 4 bits each x 4 blocks
for a total of 32 bits, this is it!! Let's call the 1st 8 bits data
block (bits D0...7) "0", the 2nd (bits D8...15) "1", the 3rd (bits
D16...23) "2" and the 4th (bits D24-31) "3".

There are 2 signals /RAS0 /RAS2 (1 per 2 blocks, hence per 4 chips)
and 4 signals /CAS0 /CAS1 /CAS2 /CAS3 (1 per block, hence per 2
chips). Address signals A0...9, plus /WE, /OE and power supplies (VSS
and VCC) are common to all chips. Signal /OE is always wired to the
ground, so it doesn't show up on the connector. The 32 data bits
D0...31 are wired by 8 (1MB block) from each chip to the connector.


***** Chipack 3240 GVP's way of working

As no drawing was given with my Combo board, I took time to track down
all signals, thanks to a simple "beeper". This way, blocks "0" to "3"
appear (each made of 2 chips 1Mbitx4) but it's like GVP had fun and
put a big mess on the connector on purpose. Look:

  * /RAS1357 to the 1st chip (odd #) of each block

  * /RAS2468 to the 2nd chip (even #) of each block

  * /CAS13 to each 1st chip of the first 2 blocks

  * /CAS24 to each 2nd chip of the first 2 blocks

  * /CAS57 to each 1st chip of the last 2 blocks

  * /CAS68 to each 2nd chip of the last 2 blocks

  * /WE12 to block "0", /WE34 to block "1", /WE56 to block "2",
    /WE78 to block "3"

  * /OE exists and goes to each memory chip

The other signals A0...9, VCC et VSS, and data bits are wired like a
"regular" SIMM.

Sidenote: There is rumor saying that these Chipack 3240 would have a
cheaper Macintosh equivalent: SIMMs for Mac II FX. After
investigation, I found out that these Apple SIMMs had the same 64 pins
connector indeed, but their size is 1MB 8 bits each only, instead of
4MB 32 bits for a GVP. Therefore they are useless.


***** GVP Combo 64 pins connector

Considering the big mess I just told you about, I was expecting for
the worst. Thankfully, many signals got connected together:

  * an only /RAS connects /RAS1357 and /RAS2468 together

    -> one /RAS for all

  * /CAS1 connects /CAS13 and /CAS24

    -> one /CAS for the first 2 blocks

  * /CAS2 connects /CAS57 and /CAS68

    -> one /CAS for the last 2 blocks

  * /OE grounded

The connection Combo -> Chipack 3240 then looks like this:


            <insert ComboChipack.IFF>


On the other hand, the 4 separate write signals /WE1, /WE2, /WE3, /WE4
still remain, although there is only one common /WE on a "regular"
SIMM. This is why the SIMM 72 pins needs to be hacked.


***** 1st step: hacking the SIMM

This one takes the shortest time but is the most delicate to make. You
will have to follow the checklist below closely:

  - Insert the SIMM into its socket, to protect it (we are going to
    SOLDER on it); be careful to insert it the right way, there is a
    little notch near pin 1 (see next drawing) to avoid inserting it the
    opposite/wrong way;

  - Find the blocks "0" to "3" thanks to their respective signals
    /CASx (x=0...3) with a "beeper", following the 72 pins connector's
    drawing closely. /CAS0 will go to block "0"*and this one only*, /CAS1
    will go to block "1" etc.


            <insert CasToBloc.IFF>


    In general, chips are physically aligned on the PCB, therefore
    blocks should do the same, but it's not sure! On the SIMM 4MB I used,
    the chips were soldered on both sides of the PCB, by pairs, so I
    easily found out the blocks #, but any kind of connection is possible
    a priori.

  - Now that you have "split" the memory module into blocks "0" to
    "3", next you need, on each chip, to CUT delicately, as close to the
    PCB as possible, each /WE leg, then lift it upwards to bend it
    horizontally (if the leg is hard to cut, use a small screwdriver and
    use it as a lever on each edge of the chip) like this:


            <insert CoupureWE.IFF>


!!! WARNING: the previous handling requires lots of precision and
EXTRA THIN tools. One bad cut and it breaks the leg apart ==> the SIMM
is destroyed and trashed!!!

  - You should now have 8 /WEx legs "bent up". Mate them by pairs (cf
    schematic).

    Pin #47 of connector 72 pins, called /WE, is now left open, since
    all /WEx legs have been individually disconnected. Check them all with
    the beeper.

  - If you chose (as I suggested) a SIMM without parity but with
    "empty" spaces, you can take advantage of these copper pads to solder
    the /WEx wires coming from each block. This solution is the easiest
    because these pads, used for the parity bits, are already wired to the
    connector. No need then to solder /WEx wires directly to the edge of
    the connector, there isn't much space left there and you risk to
    damage the SIMM socket!

    Following my advice, find (using the beeper) where the parity pins
    PE0 to PE3 go from the connector 72 pins: in principle, on 1 (or 2)
    copper pads of the "empty" chips (pin 1, or possibly 19, see drawing
    hereafter).

    Use the thin wrapping wire to connect /WEx leg from each block (hence
    the /WE of each of the 2 chips) to the copper pad PEx (either an empty
    pad or on the connector 72 pins, it's up to you):

      * block "0" : /WE 1st chip + /WE 2nd chip linked together to pad PE0

      * block "1" : /WE 1st chip + /WE 2nd chip linked together to pad PE1

      * same for blocks "2" et "3" with PE2 and PE3


            <insert WeToParité.IFF>


  - It's all set! The SIMM module must now have 4 short wires
    connecting the /WE legs of each chip, pair by pair, to a copper pad,
    linked to (or, depending on what you chose to do, directly onto) the
    connector 72 pins. Go and have some champagne, you did a great job!!


***** Get wired

It's the longest and most tedious part of the job, but thankfully less
tricky than it seems. To get everything reading for the wiring, cut
the grey cable into several 6 cm (2") long sections. Separate the
small wires from each other with an X-akto knife, into 4 chunks of 8
wires each, plus 1 chunk of 10 wires. OK OK, I know it's quite hard to
describe, please have a look at the next drawing with the cable handy
and you will understand.

Then, at each wire's end, separate it from its neighbors on a small
length (2cm or 1" max) until you get some sort of "double sided comb",
with wires still attached together by the center, like this:


            <insert Tronçons.IFF>


Then, prepare, with what is left of the cable, some longer separate
wires (10cm or 4"), for the several remaining command signals.

In the end, take the slightly thicker wire, and cut it into as many
wires (7cm or 3") as there are VCC and VSS pins on the 72-pins
connector. I advise you to use 2 different colours for VCC and VSS, so
that there won't be any short circuit to blow up everything.

All you have to do is to strip each wire over a few millimeters at
BOTH ends, then put some soldering (using the low power iron) to make
it more rigid.

You're now ready to do the wiring!


***** Connecting the 72-pins SIMM to the Combo

A drawing is self-explanatory, so here are the required connections in
every detail:


            <insert SIMM72ToCombo.IFF>


This looks rather. confused, eh? but if you follow each track from
one end to the other, there is no risk to mislead one wire for
another. I voluntarily drew some tracks thicker, otherwise they would
have been "drowned" under the rest of the wires, but there's no need
to use thicker wires for them. Note that the thickest wires are for
the power supplies VCC and VSS.

To help you, keep in mind this cross table:

Combo        Conn. 64 pins GVP    Conn. 72 pins SIMM 4Mo

-----------------------------------------------------------------

D0...7        Dx.1 (4) et Dx.2 (4)    DQ0...7

D8...15        Dx.3 (4) et Dx.4 (4)    DQ8...15

D16...23    Dx.5 (4) et Dx.6 (4)    DQ16...23

D24...31    Dx.7 (4) et Dx.8 (4)    DQ24...31

/WE1        /WE12            /WE0 (PE0)

/WE2        /WE34            /WE1 (PE1)

/WE3        /WE56            /WE2 (PE2)

/WE4        /WE78            /WE3 (PE3)

/CAS1        /CAS13 or /CAS24    /CAS0 + /CAS1 together

/CAS2        /CAS57 or /CAS68    /CAS2 + /CAS3 together

/RAS        /RAS1357 or /RAS2468    /RAS0 + /RAS2 together

/OE        /OE + pin 62 together    not connected

VSS        VSS            VSS

VCC        VCC            VCC

"Together" = solder the 2 pins TOGETHER!

/OE must return back to the Combo thru pin 62, thanks to a small wire
that you will solder directly on the Combo's PCB.

"Or" = signal coming from Combo goes to the 2 pins, so that SIMM 72
can be connected to ANY of the two (it's up to you, on the schematic I
chose for example one of the two /CAS but you freely use the other
one).


***** Soldering

I advise you to solder, in this order:

  * D0...7, D8...15, D16...23, D24...31

  * A0...9

  * /RAS /CAS1 /CAS2 /WE1...4 /OE, bridge on /OE

  * VSS VCC

The Combo connector to which the SIMM 72 is to be soldered can be
CN11, CN12 or CN13, but some basic rules are to be followed:

  * if your Combo has some RAM originally soldered on its PCB (and
    this should be the case for the first 1MB or 4MB), don't use CN10 pads!

  * solder on soldering side, not on components' side (SIMM sockets
    won't let you do that!) and pick up the NEXT FREE connector: if you
    have no Chipack installed, use CN11, otherwise use CN12 (Chipack in
    CN11) or even CN13 (2 Chipack in CN11 and CN12)

  * pin 1 on the 64-pins connector has a square-shaped pad, while the
    other pins are round-shaped; starting from the pin 1, pads on the same
    side are the odd ones (1...63) and the opposite ones are the even ones
    (2...64).

    The spacing in the center of the 64-pins connector "splits" it
    into 2 parts: pins 1 to 32 on the left (facing the Combo's
    components' side, Zorro connector on the bottom) and pins 33 to 64 on
    the right. Same for the SIMM 72-pins (pins 1...36 then 37...72). As
    expected, on soldering side, the order is reversed:


            <insert Conn64.IFF>


  * ALL VCC and VSS pins must be connected to the Combo, to make sure
    that the SIMM gets "clean" power supplies (draining enough current,
    that a wire alone can't do) .

    To the contrary, on the Combo, these VCC and VSS pads are
    connected to the power planes, the innermost copper layers in the
    thickness of the PCB. Therefore, these layers will dissipate more heat
    and a more powerful iron is required to solder these wires only.

Check carefully your connections with the beeper, between the 72-pins
SIMM connector and the 64-pins Combo connector, and even between this
one and the memory chips on the SIMM if you have any doubt.

And of course... FORGET NO WIRE!!


***** Configuration of the Combo

The important jumper not to forget is J12. Now that you have more than
the original 1MB or 4MB, you must avoid any possible conflict between
the extra RAM you just added and the FAST RAM (if any) already
installed in your Amiga on a separate Zorro expansion board (such as a
GVP HCD8+). Your Combo's manual will tell how to set J12 accordingly.

Remember: if J12 is closed (installed), there can be a maximum of 8MB
of FAST RAM (or 6MB if you own a BridgeBoard XT/AT) installed in the
system at the same time (this is the A500/2000 Zorro II limitation),
but there is no limit to the amount of "32-bits" RAM installed (on the
Combo only, with J12 open).


***** Turning it on

Hold your breath...

If the normal dark grey / light grey / white screens show up, if your
Amiga boots up, if the Workbench is back as always, you have almost
won!

You must make sure, though, that the extra 4MB have duly been found
and added to the system memory (by running "avail", "sysinfo" or
else). If not, you must have forgotten one of the many wires, or the
bridge on /OE, which seems to detect the SIMM (I didn't really check,
but it exists on the Chipack 3240 so it MUST be of some use!).

If the dark grey screen shows up but soon after the "POWER" LED starts
blinking and all you get is a yellow screen, this also happened to me:
the SIMM is found but the command signals are not connected the right
way. Did you connect all of the /RAS, /CAS and /WEx ? Is the SIMM
itself well inserted in its socket ? None of the wire got broken when
re-installing the Combo ?

In my case, I had left one of the /RAS open, then I had tried with a
normal SIMM (not "hacked") but no luck, yellow screen was back! Once
these 2 mistakes corrected, miracle, my Workbench was back on screen!


***** Footnote

I hope that reading the description above wasn't too boring.
Personally, it mainly took me 1 hour to hack the SIMM then 2+ hours to
do the wiring, LITTLE BY LITTLE. But then I duly appreciated the extra
4MB, trust me on this! I did this hack during December 1995, and it's
still working flawlessly.

Once it works for you and you are sure that you did good solderings,
do not hesitate to "drown" all the wires into glue (like Araldite) to
avoid tearing the wires away. You can also tie up the SIMM to the
Combo, using some thread (isolated of course, such as a piece of
fishing nylon thread) the way you want. No need to say that putting a
broken wire back in place isn't this easy (I had to do it).

There is a slight chance for me to design a dedicated PCB for it, to
replace these many wires... one day (bye bye spare time these days). Of
course, I will let you know, just email me some time to time!

Now a question for you: there IS some nasty trick I found and used,
about the pin-out of the SIMM, that I didn't tell you to avoid
confusing you any further! Can you see what it is ? The first one
to tell me will win a free PCB, when it will be done, that is ;->

Look out for my next articles in French magazine "AmigaNews"! Soon to
come: how to speed up Combos, add FPU, add Guru ROM, configure SCSI
etc.

Pascal "P-chan" Janin.

            <insert Pchan_Sign.iff>

EMAIL: [e-mail]