15 Apr
2002
15 Apr
'02
1:36 p.m.
True enough, but the concept was cooked up by someone in their back room or
lab back in the days when TTL was still leading edge technology, and nowadays,
the most common remark about TTL is preceded by "where can I get a ..."
Dick
----- Original Message -----
From: "Tim Walls" <tim(a)snowgoons.fsnet.co.uk>
To: <classiccmp(a)classiccmp.org>
Sent: Monday, April 15, 2002 7:19 AM
Subject: Re: TTL computing
software
From:
"Richard Erlacher" <edick(a)idcomm.com>
Date: Mon Apr 15, 2002 03:45:30 am Europe/London
piplelines
Actually, I can see one way that there could be a
benefit from
using TTL....
The current advances in processor speed have come largely from just
increasing the clock rate. There haven't been any major changes in CPU
design to use those clock cylces more efficiently....
That's only a small part of the acceleration. The use of multiple accounts for much of the performance increase
along with increased
datapath width, and other little features. The gradual increase in
interest in
parallelism is also going to help quite a bit, so
we'll be seeing even more
pipelines in the future.
> But somebody stuck with old, slow, TTL, just might hit on some way to
get
more
performance out of it (because it's all they've got, and they need
the performnce). The trick they discover just might also be useful to
speed up ASICs (or FPGAs, or ...)
And just exactly HOW would they extract more performance from it? A new
architecture would require new software, both in development tools
and in OS and as applications. Just verifying that their innovation would
take several hundred lifetimes, and the generation of a full set of would take that one individual working alone,
until well after the next
big-bang.
I've been keeping out of this conversation so far 'cos I'm not qualified
to comment on much of it <grin>, but this statement strikes me as just
being Plain Wrong.
Your own example of pipelines is a perfectly good example of
something which could easily have been invented by someone
in their back room building a processor out of TTL. OK, the resultant
processor would not be commercially viable, but it would serve
perfectly well to demonstrate the theoretical concept and a
practical embodiment of it. Enough for the patent application you'd
file before taking it to Intel to commercialise, for example!
The argument that anyone interested in new architectures is wasting
their time ('mentally masturbating' as you so delicately put it,) is
entirely specious IMHO. The first stage is always proof of concept,
commercialisation is an entirely different game. I doubt the
researchers on quantum computers are going to give up and go
home because the first devices they produce are the size of
a room, require liquid nitrogen to run and - heaven forfend - don't
run Windows 2000 or Microsoft Office. If a new architecture
offers significant enough benefits, the software support and
everything else necessary to capitalise on it will follow.
Cheers,
Tim.
--
Tim Walls at home in Croydon - Reply to tim(a)snowgoons.fsnet.co.uk