toh@micor.ocunix.on.ca (Marc Sira) writes:

>Which delays are actually necessary? I've tinkered with "CPS Follow" and
>"Counter Delay" and the like but have no idea what I'm actually doing...there
>doesn't seem to be any immediate effect to tell me, either.

CPS Follow should always be enabled. If you disable it you will have problems
with Disk ]['s and System 6's AppleTalk driver and anything else that expects
the Zip to slow down to 1 mhz when the GS is instructed to slow down to 1 mhz.
For instance, border text demos (like the FTA XMAS demo) won't work if you have
this disabled. Expect weird things to happen if you play with this one.

Counter Delay I would recommend leaving enabled. This causes the Zip to
"deaccelerate" (actually all it does is temporarily ignore that any data can
be read from the cache instead of the motherboard) for about 5 milliseconds
any time you read one of the Video Counters -- this is really only there so
that one of the self-tests will pass. Which way you set it shouldn't be fatal.

AppleTalk or Interrupt Delay should always be disabled. The desktop will run
much slower if you enable it. The only reason most people needed it was for
AppleTalk under system 5, and I have an init on tybalt that fixes that.
System 6 fixed the problem but requires CPS Follow to be enabled for the fix
to work.

Speaker and Paddle delays are intended to let old 1-bit sounds and Paddle
access work the way they do on an unaccelerated IIgs. I recommend leaving these
on unless you feel like experimenting. They shouldn't be fatal but some program
might react very negatively, so be prepared.

Bank C/D cache enable -- leave this at the default setting and forget about it.
This tells the Zip if it's OK to cache bank-switched RAM (the old language card
area). Zip's own docs say "there is no know software requiring this" but that
it why it's there, in case somebody ever finds software that doesn't like it
you can try setting it the other way.

>Can I basically mix n' match parts? That is, could I hypothetically stick in
>the new 14MHz CPU and run it with my existing crystal & SRAM?

Yes. You can always put faster CPU's or RAM's in, or at least you _should_ be
able to. Just don't expect them to actually run faster unless you change the
crystal too.

>Do I really need to buy 15ns SRAM? Particularly if I just wanna run at 12 or
>12-point-something megaHertz? Shouldn't 35ns be the correct speed for 14MHz?

That depends... since I do not have timing specs for the Sanyo chip, I have no
idea what speed it would really require. For a while the rule of thumb with
the WDC parts was half the clock period of the CPU, but that isn't true of the
higher (?) speed WDC parts. From what I remember of the data sheets WDC sent
me once, some of their higher speed parts were nearly the same, but required
everything else to be faster -- which is somewhat bogus.

I would be willing to bet that 25ns would be enough. 35 seems a little chancy
but since I have not actually gotten my hands on any of this I am far from an
authority on the new chip. Remember that the Zip cache controller ASIC has a
delay going through it and this is on the order of 10-15 ns (according to what
the Zip tech dude told me). As the CPU gets faster, that overhead gets more
significant. That is why I think the 1/2 rule breaks down.

>And does anyone have a company where we aspiring hobbyists can order wide,
>600 mils SRAM at these speeds? And part numbers?

Somebody else on the net has posted this info before. If anyone has a copy,
please post it. 

>ditto for the oscillator crystal...wherefore a 48 or 56MHz crystal?

Pick the speed you want to max out at. The crystal should be 4x that. If the
CPU & RAM (and the Zip ASIC) are fast enough to work at that speed, you're
happy. Otherwise you will get flakiness or blatant crashes depending on how
close you are to the edge, and how warm the computer is, and other moon phase
related things.

>Finally, what are the most likely ways of actually damaging the board with this
>sort of experimentation? Are there any?

When you run a chip too fast it does two things.

	1. it screws up.
	2. it gets warmer.

#1 is why the chip is rated at the speed that's stamped on it. Stuff inside
the chip happens at a finite speed which is dependent on a lot of things, so
they test every chip to see what speed it is safe to run at (and then some, to
anticipate normal wear and tear over the expected lifetime) and that's what the
speed rating is. It is very difficult for a chip to get physically damaged if
it is run too fast; it will just get -very- confused and you won't be able to
make it do anything coherent until you run it at a safer speed.

#2 is a simple consequence of electronics principles and CMOS -- the current
used by a CMOS chip is roughly proportional to the clock speed you run it at,
so naturally the amount of heat it's going to give off will increase as you
force it to use more current (conservation of energy, folks, the stuff has to
go somewhere). The kind of chips you're using in a Zip won't ever be in danger
of overheating themselves; this is only a problem with the mega-transistor CPUs
like the early 486's that were burning up on the test pad.

Todd Whitesel
toddpw @ cco.caltech.edu