18 Apr
2005
18 Apr
'05
10:05 a.m.
Subject: Re: PDP11/55 sells on Ebay for 5K$ - was it
really the fastest 11?
From: "Heinz Wolter" <h.wolter at sympatico.ca>
Date: Mon, 18 Apr 2005 01:00:13 -0400
To: "Lyle Bickley" <lbickley at bickleywest.com>, <cctalk at
classiccmp.org>
snipe
Lyle wrote:
McNamara),
Sorry, while for those metric the 55 was a tad faster, for IO the 11/70 was
massively faster. At that time to do large arrays of data you needed lots
of fast IO to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
When you measure systems, measure the system not just the cpu.
Allison
> I expected an 11/70 to be instruction for
instruction faster - but
> add all the cache and complexity - maybe the 55 with bipolar ~was~
faster?
How about
compared to a 11/70 with a PEP70 instead of factory memory and
cache?
Isn't a 55 just a 45 with bipolar instead of core? or was that a 50?
Read "Computer Engineering" (Gorden Bell, J. Craig Mudge, John E.
Copyright 1978 by Digital Equipment Corporation
Page 408
--------
Model Basic Instructions Floating Point
Inst. per second* Inst. per second
-------------------------------------------
11/70 36 671
11/55 41 725
* Relative to 11/03
No brainer - the 11/55 wins hands down.
sigh - I haven't memorized the DEC sacred texts yet;) no hardcopy, only
pdf...
So the 55 is roughly 14% faster than 70 - probably due to main memory
timing.
This might not be the case with the PEP70, though it's not stock.
Interesting
that the FP is only 8% faster - given that they are likely similarly
microded and
memory would not be such an issue working out of the FP registers. Clock
speed?
Then again... if one put 2 MW of 300ns ram on a 70.. well the 55 was only
18 bit address ;)
Anybody know the microclock frequency of both models? Variable timing/phase?
This variable microclock trick was later used in Vax/750 hot wiring kits..
On a similar vein, wasn't the 11/34c clock slowed to allow for cache timing,
making
it slower than a 34A? (for uncached accesses) Is is this a possible reason
for the 70's
poor performance? Running code withing the cache should have beaten the 55!
h 
18 Apr
18 Apr
10:41 a.m.
On Monday 18 April 2005 07:05, Allison wrote:
--snip--
>> Read "Computer Engineering" (Gorden
Bell, J. Craig Mudge, John E.
>
>McNamara),
>
>> Copyright 1978 by Digital Equipment Corporation
>> Page 408
>> --------
>> Model Basic Instructions Floating Point
>> Inst. per second* Inst. per second
>> -------------------------------------------
>> 11/70 36 671
>> 11/55 41 725
>> * Relative to 11/03
>> No brainer - the 11/55 wins hands down.
Sorry, while for those metric the 55 was a tad faster,
for IO the 11/70 was
massively faster. At that time to do large arrays of data you needed lots
of fast IO to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
When you measure systems, measure the system not just the cpu.
System performance is application dependent.
If you were going to do FORTRAN FP, the 11/55 would be a good choice. This
specific 11/55 it was used as part of a flight simulator - a perfect
application for the high performance (CPU) 11/55.
If you were running a RSTS/E shop with lots of I/O (as I did many years ago),
your choice would have been an 11/70 - for it's massbus I/O capabilities,
good integer/FP performance, and memory capacity.
So saying it is the "fastest blinkenlights" system is certainly valid in the
context of its application - and in the manner that "Computer Engineering"
documented its performance.
Cheers,
Lyle
--
Lyle Bickley
Bickley Consulting West Inc.
http://bickleywest.com
"Black holes are where God is dividing by zero"
3:42 p.m.
Lyle Bickley wrote:
>Copyright 1978 by Digital Equipment Corporation
>Page 408
>--------
>Model Basic Instructions Floating Point
> Inst. per second* Inst. per second
>-------------------------------------------
>11/70 36 671
>11/55 41 725
>* Relative to 11/03
>No brainer - the 11/55 wins hands down.
>
Sorry, while for those metric the 55 was a tad faster, for IO the 11/70 was
massively faster. At that time to do large arrays of data you needed lots
of fast IO to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
When you measure systems, measure the system not just the cpu.
1:41 p.m.
Allison wrote:
Sorry, while for those metric the 55 was a tad faster,
for IO the 11/70 was
massively faster. At that time to do large arrays of data you needed lots
of fast IO to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
Umm how about real HARD Numbers when talking speeds.
When you measure systems, measure the system not just
the cpu.
Or like everbody does today clock speeds ... with a .3 ns clock
how fast is
REAL memory again?
Ben alias woodelf
2:04 p.m.
On Mon, 2005-04-18 at 11:41 -0600, woodelf wrote:
--
TTFN - Guy
Allison wrote:
The basic instruction timing on the 11/45, 11/50, 11/55 and 11/70 is
300ns (ie the time it takes for the *fastest* instruction to execute).
There are adders for the different types of memory. On the 11/45,
11/50, 11/55 there is a 0ns penalty for bipolar. It gets slower from
there. I don't have the 11/70 docs in front of me to determine the slow
downs for an 11/70.
One of the other reasons that the 11/45, 11/50, 11/55's are fast when
you put memory in the fastbus is that the fastbus memory is dual ported
so the CPU doesn't slow down when there are other masters on the unibus
(actually they have 2 unibuses but in most applications they are tied
together).
Sorry, while for those metric the 55 was a tad
faster, for IO the 11/70 was
massively faster. At that time to do large arrays of data you needed lots
of fast IO to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
Umm how about real HARD Numbers when talking speeds. When you measure systems, measure the system not
just the cpu.
Or like everbody does today clock speeds ... with a .3 ns clock
how fast is
REAL memory again?
Ben alias woodelf
2:28 p.m.
At 11:04 AM 4/18/2005 -0700, Guy Sotomayor wrote:
--
TTFN - Guy
On Mon, 2005-04-18 at 11:41 -0600, woodelf wrote:
The basic instruction timing on the 11/45, 11/50, 11/55 and 11/70 is
300ns (ie the time it takes for the *fastest* instruction to execute).
There are adders for the different types of memory. On the 11/45,
11/50, 11/55 there is a 0ns penalty for bipolar. It gets slower from
there. I don't have the 11/70 docs in front of me to determine the slow
downs for an 11/70.
One of the other reasons that the 11/45, 11/50, 11/55's are fast when
you put memory in the fastbus is that the fastbus memory is dual ported
so the CPU doesn't slow down when there are other masters on the unibus
(actually they have 2 unibuses but in most applications they are tied
together).
I believe that speed of access to the Unibus on the 11/45 (et al) is also
quite a bit faster than that on the 11/70, although I don't recall the
reasons.
--tom
Allison wrote:
>Sorry, while for those metric the 55 was a tad faster, for IO the
11/70 was
>massively faster. At that time to do large
arrays of data you needed
lots
of fast IO
to disks as you could only works with part of an array at any
time due too addressing limitations of the PDP11.
Umm how about real HARD Numbers when talking speeds. When you measure systems, measure the system not
just the cpu.
Or like everbody does today clock speeds ... with a .3 ns clock
how fast is
REAL memory again?
Ben alias woodelf
7035
days inactive
7035
days old
5 comments
6 participants
participants (6)
-
ajp166＠bellatlantic.net -
bfranchuk＠jetnet.ab.ca -
ggs＠shiresoft.com -
jhfinexgs2＠compsys.to -
lbickley＠bickleywest.com -
uban＠ubanproductions.com