> From: Philip.Belben_at_powertech.co.uk
> Subject: Re[2]: Talk Of Building A Computer...
(Don't react until you read this first part through:)
You can of course never build a real classic computer, by definition its
not a real classic unless its really old and was built by people who
were, at the time, working with state of the art components and
techniques. Part of the attraction of the old machines is that they
subtly document the skills, preconceptions and ignorance of their
designers and the prevailing conventional wisdom. Another part of the
attraction is to have a piece of equipment that has been in existence
for so long and still works.
So the best you could ever do is come close. Any attempt will be a
compromise of some sort, so an interesting question is how close is
close enough? The closest I think is to attempt to exactly recreate a
particular early machine. Chris Burton is doing this with the Manchester
SSEM and Tony Sale with the Colossus, both with significant help from
other folks in the Computer Conservation Society. I've seen them in
progress, they are both fabulously, meticulously accurate (as far as I
can tell) and very nearly as much fun as if the original machines still
existed. And despite all the help they are largely one-man projects, or
could have been, which shows that its actually possible to consider
doing it "right" by yourself if you have the time. Sort of like building
a boat or an airplane. People can do that.
All this having been said, I'm sure you feel as I do that it would be
fun to build a machine in the classic style but an exact recreation is
too much. So how authentic do you have to be? My point is this -
*** Its up to you ***.
Since any such project is necessarily a compromise, the exact tradeoffs
are not really important. Your project should reflect what you want from
it, other people's opinion doesn't matter unless its a group effort.
Here are a some projects I personnally find interesting:
1. A small Williams-tube memory. Designed and built from scratch using
6SN7's and a common 5" oscilloscope CRT like maybe a 5GP1.
(Chris has beaten me to this but it doesn't matter - for that matter,
Williams beat us both; the fun part is doing it yourself.)
2. A complete 32-bit CPU and memory in the classic style, an accumulator
machine, built in just 4 parts: a FPGA, DRAM, EPROM and clock. This is
easily possible with existing technology. It would also have a 2.5" IDE
disk drive for mass storage and a small printer and keyboard for user
I/O. It would be fun to do and would demonstrate the incredible
miniaturization of electronics when compared with my room sized IBM
7094.
3. Emulators for all my old machines, and machines I wish I had. This is
the only project I've actually made any progress on.
So do what you find most compelling.
--------------------------------------------------------------------
Here are some thoughts on the current proposal (building a machine in
the classic style of a tube machine but with discrete transistors to be
more practical), some of these reinforce comments others have already
made:
Your main problem is not the logic, it is the memory. There is no old
memory that is anywhere near as easy to use as, say, SRAM. Your major
choices are core, magnetic drum/disk, or delay line. Core is going to be
a lot of work. Drum or disk will require expertise in mechanical
fabrication but is pretty attractive otherwise. Acoustic delay line (the
only kind with reasonable capacity) may be best but will require some
research and experimentation. I would recommend magnetostrictive wire
acoustic delay line memory if you can figure out how to build it and
don't mind having a very small memory.
About the logic, while very early tube circuits were strange using
multiple grids, in most tube computer circuits the actual logic was done
with diodes. The tubes then invert and drive the next stage. Flip-flops
were used for temporary state storage and sometimes for registers. A
very common computer tube in the US was the 5965 dual triode. To see
some IBM 705 circuit drawings (all but the inverter, which apparently
wasn't used much in this machine but was used in others such as the 709)
look at
http://www.teleport.com/~prp/collect/705dwg/
The diode/tube logic translates very nicely into common early transistor
circuits such as RTL, RCTL and DTL.
For a project like this I would recommend building a serial accumulator
machine, one which works on one bit at a time and has a single
architecturally visible register, like the PDP-8. It will have much less
logic than a parallel or multiple register machine and will be very
classic. A serial CPU will be a good match with serial memory such as
drum or delay line memory. A good, pretty clean example is the Royal
McBee LGP-30. Better is the SWAC which had a very good clean minimal
architecture but was parallel, a serial version would be pretty easy.
Note that small tube machines like the LGP-30 and Bendix G-15 have only
around 300 tubes.
To get an idea of how big it would be, the Packard Bell PB250 (see
http://www.teleport.com/~prp/collect/mini.html ) is just what I've been
talking about - serial, transistors (RCTL) and delay line memory. It
takes up half a 19" rack, same as a PDP-8 but I think the logic is
physically less dense. About a quarter of the volume is taken up by the
memory.
Paul
Received on Fri Nov 21 1997 - 12:42:14 GMT