>Pete Turnbull wrote:
>No. There's one board, with one number. Those sold as 11/73 had
>processors running at 15MHz and didn't have the FPU fitted as standard,
>those sold as 11/83 or 11/84 had faster processors (or at least a
>faster clock) and did have an FPU. The boards themselves are the same.
>
>
Jerome Fine replies:
After attempting to check the KDJ11-BF board, I now
agree.
From what I heard, when DEC originally attempted to
use the 18 MHz crystal, the FPU bug was not initially
apparent and it took a while before DEC realized that
the higher speed required a change to the micro code
for both the CPU and the FPU. But it was possible
to reliably run the older versions of the CPU chip with
the FPU using the 15 MHz crystal.
> You might be thinking of the M8192 KDJ11-A which is a smaller board
>used mainly in OEM equipment and for 11/23 upgrades, or perhaps of the
>11/53 or 11/93 boards.
>
No confusion there. It was just that when I originally
managed to see the "AE" module version, I could not
see the M8190 module number. Now that I realize
that all PDP-11/73,83 quad boards are M8190-Ax,
my memory has been helped to remember that you
are correct.
>>What is not well known (or at least not in the beginning)
>>was that ALL of the KDJ11-Bx boards could run with or
>>without PMI memory. In addition, even if PMI memory was
>>used, if located BELOW the CPU board, it was used as
>>ordinary memory, if used ABOVE the CPU board, it was
>>used as PMI memory.
>>
>Indeed. Diagnostics and operating systems distinguish 11/73 from 11/83
>by looking at a bit that relates to PMI memory.
>
>
I think you are referring to what the software sees when the PMI
memory is either active (above the CPU) or not (below the CPU).
However, I have no understanding about what the hardware does to
activate the PMI stuff when the memory is above the CPU.
Perhaps you might explain that and why PMI is faster. ALSO how
the I/O devices are able to use the memory in the normal manner
without interfering with the CPU using the memory at the same time.
>>MOST of the increase in speed for a PDP-11/83
>>was due to the PMI memory, not the shift to 18 MHz.
>>
>Well, about half, depending on the makeup of the benchmark used. I
>tried the permutations once, by re-ordering boards and changing the
>crystals (what I was really trying to do was see how fast I could get
>one to go: 20MHz with the chips I had).
>
The benchmark I used was mainly to use the VMX.SYS
device driver under RT-11. A KDJ11-BF board (with the
18 MHz crystal and PMI memory above the CPU) was about
33% faster than a KDJ11-AB board ( with the 15 MHz
crystal with PMI memory below the CPU).
While a PMI memory board below the CPU was a bit slower
than a Plessey non-PMI memory board (also below the CPU),
that difference was not very large.
What I did find was that going from 15 MHz to 18 MHz
produced about a 13% decrease in the time to execute
the benchmark. Going from using the PMI memory below
to above produced about a 20% decrease in time (when
the 15 Mhz crystal was being used). When I went from
15 MHz to 18 MHz and placed the PMI memory above the
CPU, the benchmark time decreased by about 33%.
On the other hand, I also seem to remember that the same
benchmark showed about a 10% decrease in time going from
a PDP-11/83 (M8190-BF with PMI memory above the CPU) to
a PDP-11/93 (with the memory on board). I concluded from
the 5 different tests that the PDP-11/93 was vastly over-
priced and over-rated relative to a PDP-11/83 EXCEPT for
the fact that a 4 MByte PDP-11/93 uses a single quad slot
whereas the equivalent 4 MByte PDP-11/83 needs 4 quad slots.
For a system that needs a large number of boards, the extra
3 boards required by the PDP-11/83 could be so difficult
to solve that the PDP-11/93 is justified.
BUT, I do have a problem which I have never been able to
solve. I use the PDP-11/83 with 2 PMI memory boards inside
a BA123 box. I can't seem to remember having a problem
when the EDSI controller (Sigma RQD11-EC quad ESDI controller
with 3 Hitachi DK515-78 600 MByte ESDI hard disk drives)
was initially used. BUT, about 6 years ago, I found that
if there was more than 7 quad boards in the BA123 (or
their equivalent in dual boards), then the time to read
a large number of blocks from the Hitachi hard disk drives
just about doubled. This meant that when I wanted to compare
a 32 MByte RT-11 partition on one Hitachi hard disk drive
with the same 32 MByte RT-11 partition on the second Hitachi
hard disk drive (in order to verify that there were indeed
identical), the total time doubled from about 4 minutes to
about 8 minutes. As long as the total number of quad boards
is kept to 7 quads or less, I am presently OK. Since I
no longer use that system very much, especially when I
need to use more than 7 quad boards, I have given up
trying to solve the problem. Do you have any suggestions?
I seem to remember that writing to the Hitachi disk drive
is not a problem, only reading! I also seem to remember
that when I used a Maxtor XT8760E (of the same block
capacity), the reads were OK. I tend to think that the
problem is because the sector size is a bit too small -
the Hitachi drive allows only two sector sizes close to
600 bytes and even the larger option may be too small whereas
with the Maxtor XT8760E, the sector size can be set to
a value which is about a dozen bytes larger. In addition,
the number of sectors per track with the Hitachi hard
drive is much higher than for the Maxtor XT8760E hard drive
since the Hitachi DK515-78 hard drive has many fewer total
cylinders for the same total byte capacity.
Sincerely yours,
Jerome Fine
--
If you attempted to send a reply and the original e-mail
address has been discontinued due a high volume of junk
e-mail, then the semi-permanent e-mail address can be
obtained by replacing the four characters preceding the
'at' with the four digits of the current year.
Received on Sun Feb 06 2005 - 21:17:25 GMT