You missed my point, Allison. What I hoped you'd get from that message was
that there were MANY parts involved in even a small memory array, which
meant a longer decoder tree than what we'd use today, or even 15 years ago.
Since the specified prop-delay is on the order of 35 ns, multiplied by two
or three, the difference between the TTL parts at nominally 25 extra ns of
prop delay per decoder, the prop-delay difference is significant, and it's
especially so in the context of the slow memory devices of the time. If
there were only a 250 ns access time for memory devices, it would have been
insignificant because it consumed a small part of a large window. In this
case, however it consumed a large part of what was left of the window after
the access time was deducted from the window.
I've embedded a few remarks below, too.
Dick
----- Original Message -----
From: <allisonp_at_world.std.com>
To: <classiccmp_at_classiccmp.org>
Sent: Wednesday, May 24, 2000 6:45 AM
Subject: Re: Re[2]: Altair parts substitutions
> > select a bunch of memories, and one to select a bunch of '138's that
> > selected the memories. Back when small (1K-bit) memories were the most
> > common sort, there were lots of chips expecting to see that select
strobe,
> > and the worst case load made the prop-delay long.
>
> Compared to the typical 500-650ns prop delay of common memories back in
> 1976 teh prop delay of the '138 was relatively short.
>
>
> > configuration might have been a bank of 72 2102's (they liked parity
back
> > then) and about eight 1702's. One 138 would drive each of the rows of
RAMs,
>
> I found it to be rare except in mainframes and then it was usually 4kx1
> parts which were common and available.
>
In '77 I read one memorable article in which an IBM exec was complaining
that they couldn't get enough of the 4kx1 i2147's (55 ns, 70 ns, 120 ns) to
meet their needs. It was a couple of years yet, before we small users could
even get the TI 4044, NS 5257, or i2141. They were still pretty costly in
'79.
>
> > layout I suggested. 2102's back then had a typical access time of 600
ns
> > and 1702's were somewhere between 750 and 1000 nanoseconds. I don't
think
> > one could claim that 50 ns is negligible with a 2 MHz 8080 under those
> > conditions.
>
> It was. The 2102s were fast enough to run with no waits but 1702s were so
> slow another 50ns was nothing as waits were added in 500ns increments.
>
That, precisely was the critical factor. Nobody wanted to run with more
waits than absolutely necessary.
>
> By 1977 the 1702 was long gone and nearly history for anything other than
> hobby and existing production and the 2508/2716 were moving in fast
> with a 450ns Tacc.
>
Yes, but ... the 5-volt parts were costing on the order of $100 per piece.
I bought a bunch for a project and still remember the "sticker-shock,"
though it wasn't any better a year later when I had to buy 2732's. In the
latter case, I sold them for $25 each to a west-coast surplus vendor who
turned around and sold them for $80 each.
>
In '77, Intel's boards were shipped populated with 2708's. I bought lots of
their boards and even more of the EPROMS. 2102's were typically 600 ns,
while the 21L02's, which were easier on the power budget, were quite
expensive by comparison. This had changed significantly by late '78, as had
the TTL market, but 450ns 21L02's weren't the fastest available. They were
just what Intel shipped. That year Intel started shipping their 8020-4
boards which were the first ones I got with the 8080-A and 2708's in
combination with 2114's. These were 450 ns parts. Within a year, they had
shifted to an 8085 board numberd iSBC-8024 which used 8708's (just 5-volt
2708's) and those really silly 8185's which were srams with multiplexed
address/data inputs compatible with the 8085's timing. By that time I
decided to quit using Intel SBC's.
>
> Allison
>
>
Received on Wed May 24 2000 - 10:23:14 BST