64bit data 128Bit address...Re: Building a Z-80 (64bit!!!)

From: Arfon Gryffydd <ArfonRG_at_DIEspammerSCUM.Texas.Net>
Date: Fri Oct 23 12:29:00 1998

>Just tell me why ? Because it has been the tame with old
>8 Bit uPs ? Come on thats just the stupid 'thats the way
>it always has been' idea. Address bus shuld be a) as wide
>as needed and b) if possible not exeeding the Size of the
>basic data unit.

Again I say, You are not thinking about the future! "As wide as needed"...
 Is a very NOW statement.



>For a) just tell me where you need more
>than 4 Gig, or better where you need _way_ more than 4 Gig,
>because just 4 or 10 times this address space is easyer done
>with paging methods (wait for b)).

Wait for Windows 2003! I'm sure that'll take at least 1Gb. And MS Office
2006 Hell, there's another 1 Gb. (Yes, I gotta poke at MS)


>And for b) you need
>special logic to perform address calculation, you can't
>hold one address in one regular register, you need more
>silicon to implement the adressing logic, adress calculation
>and (hidden) adressing register - you also need an additional
>ALU.
More Silicon!?!? Where has this discussion gone awry???? I wanted to
build a board the that was a Processor unit that talked Z-80. As for the
notion of making a chip.... The chip size would not change. Yes the used
real-estate would but you have that real-estate anyway.


>> If you keep thinking like "you'll need $16,000,000,000.00
>> in memory for that address size" you'll do the same stupid thing as:
>
>If you don't need it don't spent the silicon.

Again, the chip size would be about the same. As for $16,000,000,000.00
for the memory for a 128bit address size.... EASY!!! Don't add that
much!!!! But don't make it so you can't add it in the future if we start
getting 1Gb memory modules.


>> Micro$oft ("640K is more than enough for anyone"),
>
>Be carefull, now you're entering historical teretorry.
>MS-DOs was designed to use the full 1 Meg address space
>of the 8086/88. And could could well boot with something
>like 900 and more K user mem available. The 'Barrier'
>(640 or what ever) was given to DOS by the BIOS of the
>respective machine. And the PC-Bios had just to obey the
>CGA memory.

Micro$oft/PC BIOS..... It doesn't matter! the point was that someone said
"640K is more than enough for anyone" and we all got bit by it.


>> old programmers ("year 2000 is 20 years away.
>> This stuff will be obsolete by then"),
>
>You are just talking to one of them - and FYI we continue
>to use systems with two digit years and they are Y2K ready :)

Cool! I'm just goint to set my machines back to a year that matches our
calendar.



>> and let us not forget Intel with their screwy memory addressing
>> schemes on their pre-X386s.
>
>Again, please think a bit - the x86 segmentation is a real
>great atempt to give
>- give more than 64 K Mem to a 16 Bit uP

If Intel had used 32bit address lines for a 16bit uP, they wouldn't have a
64K problem. KISS!


>> Think ahead! It wouldn't cost that much more or requrie much
>> more effort to add a few bits to the address bus.
>
>Just remember b) you NEED more effort, especialy things
>that don't fit into your standard model.
>And if your thinking about MMUs to expand a 64 Bit adress
>to 128, you just cut the adressrage down to 64 Bit, since
>external address enhancements are no design detail of a CPU.

You need more effort if you want a fairly un-buggy design several
generations from now. Patches are NOT the answer. Planning is.

Magjor thought here: IF you have more than one data bus width on your
address bus, you will have to have a partial address register for the MSBs.
 This is insane! Later in uP models, your old processor's address space
AA44H will be the same as AB44H will be the same as CC44H (as an example).
Case in point the older Intel x86s. Some screwy software developers will
not take expansion into account and they will write code that will be buggy
later in this families' life.

Add to the cons the added headache of writing software that drops the MSBs
of the upper address areas or manipulates the addresses that are not a
multiple of the data word sizes.


>> As for the suggestion of using a DEC alpha to emulate a Z-80... I thought
>> about it but, those darned Alphas are way too expensive for us hobbist.
>
>Please ? An Alpha board with CPU is only a bit more expensive
>than a PC board with an equal Pentium. At least thats the
>street price over here in Munich - we have some PC part shops
>that are also selling Alpha boards and CPUs right aut of the
>blister.

Whoa!!! I want an Alpha!!!! I hope they have them in the $99.99 range!
I hate to pay a lot of money for something that will become obsolete and/or
drop like a rock in price in a few months. (I NEVER buy the latest and
greatest! The price vs. performance ratio is not acceptable in my opinion).


>> Anyone know if Zilog is going to beef up the Zx80 line????
>
>Faster 180s and of course the 380s. But Zilog has to
>build things that will sell, so they are more into
>integrating special I/O devices than more CPU power.

When I talked to Zilog (about a 2 years ago) they had no real plans to make
the Z-x80 lines viable as anythings other than controllers. Tis ashame. I
really wish they (or another company) would put the Z-x80 family in the
fight as a first class uProcessor. Get them clock freqs up! Get those
data paths wider! Get parallel processing going inside the package!

I really want to do Video and massive amount of data transfers with Z-x80s.

Arfon
"Programming today is a race between software engineers striving to build
bigger and better idiot-proof programs, and the Universe trying to produce
bigger and better idiots. So far, the Universe is winning." -- Rich Cook
Received on Fri Oct 23 1998 - 12:29:00 BST