5 Oct
2007
5 Oct
'07
10:28 a.m.
Will wrote:
Someone wrote:
> VAX9000 built of ECL100K, fastest of the fast. The second most common
> use of TTL was in very high speed instrumentation and specifically frequency
> counters and UHF PLLs.
If the 9000 series was till using 100K, DEC must have
been sleeping.
By the 9000s development period, ECL was beyond 100K. I am not sure,
but 10E may have been out by then. 10G maybe as well, but using that
leads to insanity.
Using ultra-fast ECL doesn't make much sense when you've got nanoseconds
of delay to the backplane, to the next board, and back to the part that
needs the signal.
The ECL technology used in the VAX9000 was gate arrays with roughly the
same timing parameters as 100K ECL (0.5 to 1.0 ns propogation delays).
But one thing I notice about DEC's ECL machines
(9000, KL10) - for
being ECL, there sure were ssssllloooowwwww.
KL10 was 100 series 10K ECL technology, typically 3 to 4 ns propgation
delay. A lot easier to build large systems with than say 74F00 stuff but
not a whole lot faster.
Responsiveness of a computer system depends on a lot more than the
speed of the semiconductors used to build it. Plenty of modern examples
of how to make fast silicon seem slow are coming out of Redmond I
notice :-).
Tim.
5 Oct
5 Oct
10:42 a.m.
Tim Shoppa wrote:
KL10 was 100 series 10K ECL technology, typically 3 to
4 ns propgation
delay. A lot easier to build large systems with than say 74F00 stuff but
not a whole lot faster.
I guess we need a 36 bit computer built with 74Fxx. :)
Responsiveness of a computer system depends on a lot
more than the
speed of the semiconductors used to build it. Plenty of modern examples
of how to make fast silicon seem slow are coming out of Redmond I
notice :-).
But how fast is the silicon anyhow? You read the fine print after all
the cache has loaded.
Tim.
Could the PDP-10 instruction set have been
simplified for a faster
machine. Mind you I think you would still be stuck with 18bit address
because of LISP and having two addresses in one word.
Ben.
10:57 a.m.
Could the PDP-10 instruction set have been simplified
for a faster
machine. Mind you I think you would still be stuck with 18bit address
because of LISP and having two addresses in one word.
Of course, any architecture could get a speed boost when you start
pruning out those pesky instructions.
PDP-10 was pretty clean, actually.
--
Will
11:48 a.m.
woodelf <bfranchuk at jetnet.ab.ca> wrote:
Could the PDP-10 instruction set have been simplified
for a faster
machine. Mind you I think you would still be stuck with 18bit address
because of LISP and having two addresses in one word.
I was at DEC when the KC10 (aka "Jupiter") was under developement, and
died, working in the FORTRAN-10/20 group on a compiler revision to
generate code for "Extended Addressing" mode (see more below).
I seem to recall being at a meeting where the Jupiter folks were
giving us a talk, and it comming out that indirect addressing [recall
the PDP-6/10 had one bit in each instruction for indirect, which
meant: calculate the effective address using the index register (if
non-zero) and the right 18 bits of the instruction, and then fetch the
indicated word, and start all over, and if the indirectly fetched word
has the indirect bit lit, you start again....] going to be slow,
which was a serious problem, since one important way to access memory
outside your current 18-bit "section" was to indirect thru a word with
the high order bit set. The extended addressing design allowed for 30
bit virtual addresses -- 12 extended address bits, or 4096 18-bit
sections. The KL only implemented 7 bits of extended addressing (16
sections), but the Toad-1 implemented all of them, requiring addition
work on the monitor, and, I think another layer of indirection in
section/page mapping.
In hindsight, I'm not sure why we couldn't have "made do" using
"base
registers" ala S/360 for objects in other sections or performing
addressing arithmetic explicitly, and then indexing, unless indexing
couldn't reference data outside the "local" section.
William Donzelli <wdonzelli at gmail.com> wrote:
Of course, any architecture could get a speed boost
when you start
pruning out those pesky instructions.
PDP-10 was pretty clean, actually.
Except for the EXTEND instruction, which had all kinds of goop
including MOVSLJ (move-sludge) a string copy instruction and kinds of
stuff only a COBOL runtime implementor could love....
phil
10:55 a.m.
Using ultra-fast ECL doesn't make much sense when
you've got nanoseconds
of delay to the backplane, to the next board, and back to the part that
needs the signal.
That depends on how tight everything is. Thermal Conduction Module, anyone?
The ECL technology used in the VAX9000 was gate arrays
with roughly the
same timing parameters as 100K ECL (0.5 to 1.0 ns propogation delays).
Yes, but I do not think that was the cutting edge anymore. Considering
the 9000 was supposed to be the machine that finally convinces the
mainframe world to accept DEC, it may have been a poor choice. We
probably will never know. 9000 may have been as big of an
embarrassment as the KC10.
Even though the 9000s were bombs, they are one of the few VAX machines
I would chase after.
Responsiveness of a computer system depends on a lot
more than the
speed of the semiconductors used to build it. Plenty of modern examples
of how to make fast silicon seem slow are coming out of Redmond I
notice :-).
I am thinking raw horsepower - all the benchmarking stuff. Looking at
the KL10 (or the other DEC ECL machines), it justs seems like they
should have been better number crunchers.
--
Will
2:28 p.m.
On 5 Oct 2007 at 13:28, Tim Shoppa wrote:
Responsiveness of a computer system depends on a lot
more than the
speed of the semiconductors used to build it. Plenty of modern examples
of how to make fast silicon seem slow are coming out of Redmond I
notice :-).
I seem to recall that there was a Honeywell mainframe back in the
70's that was completely re-engineered to make use of ECL--with the
result that the new machine wasn't much faster than the older one.
Anyone recall the system? Was that the one with direct water cooling-
-and algae in the cooling lines?
Cheers,
Chuck
6240
days inactive
6240
days old
5 comments
5 participants
participants (5)
-
bfranchuk＠jetnet.ab.ca -
cclist＠sydex.com -
phil＠ultimate.com -
tshoppa＠wmata.com -
wdonzelli＠gmail.com