• Please review our updated Terms and Rules here

"simple" question for PDP 11/45 and 11/70 experts - location of memory boards

Assuming that it's Unibus-based memory, then anywhere on the Unibus will do. It's generally desirable to place it close to the CPU in the main cabinet, but there may be reasons to put it in a secondary cabinet (space, power, etc.). The 11/70 almost always uses a specialty memory unit (MJ11 and/or MK11) which typically ends up in a secondary cabinet but that's mainly because a typical installation hosts multiple of those 6U units. The 11/70 doesn't have many Unibus slots free in the main chassis so if Unibus memory is used then it's very likely to be in a secondary Unibus expansion chassis (e.g., BA11-K), but perhaps still in the main cabinet.
 
Certainly on the 11/45 I would be surprised if there was not memory slots in the CPU chassis. However, the slots themselves (and the associated power supplies) will be 'tailored' for a specific set of memory card types and configuration.

Dave
 
Paul correctly notes that with Unibus, you can put the memory "wherever you like" physically and it will "work" assuming you configure it correctly, with one major caveat. Read the manuals carefully! The caveat is that for "fast" semiconductor memory on the 11/45, there is a dedicated set of slots in the CPU cabinet on a "second" Unibus (called FASTBUS in the manuals). If you have core memory you can't use this second Unibus, and you aren't required to use it in any event. (I am not sure about the 11/70; I only have an 11/45.)

At least for core memory, placing the memory in different places has performance implications, most notably:
  1. The interrupt grant lines chain through devices in sequence, so devices nearer to the CPU will tend to get priority, and
  2. Long Unibus lines can end up with signal degradation (ringing, etc.) that can impact performance.
Regardless, the memory slots are usually custom backplane modules that can be moved around at will. It seems that the factory configuration for the 11/45 usually had the memory installed in the CPU cabinet. You must be careful to make sure that the correct boards go in the correct backplane modules and in the correct slots within these, otherwise disaster will swiftly follow!

For many years, I ran my 11/45 with core memory in a cabinet that was separate from the CPU. I did this because I picked my machine off a literal trash heap, after someone had stripped the power supplies. Not wanting to try to find original power supplies, I chose to build modern switching power supplies instead. (Read: *much* lighter!!!) The +20V power supply needed to run the core memory has harder for me to build than the other voltages, so my CPU cabinet didn't have +20V available. I found another chassis with all its power supplies intact, including the +20V. Putting the core memory in that cabinet and running a long-ish Unibus cable between the two cabinets did get me a working machine. I've since rebuilt the CPU cabinet power supplies, including the +20V, and so have moved the core module and boards back into the CPU cabinet. I don't notice much of a difference in speed, in part because core is rather slow anyway. But the extra space in the other cabinet turned out to be fortuitous: it now houses a pdp-11/34.
 
Michael Robinson said:
Paul correctly notes that with Unibus, you can put the memory "wherever you like" physically and it will "work" assuming you configure it correctly, with one major caveat. Read the manuals carefully! The caveat is that for "fast" semiconductor memory on the 11/45, there is a dedicated set of slots in the CPU cabinet on a "second" Unibus (called FASTBUS in the manuals). If you have core memory you can't use this second Unibus, and you aren't required to use it in any event. (I am not sure about the 11/70; I only have an 11/45.)
Click to expand...
AFAIK the only Unibus-based CPU that could be said to have two of them was the 11/60; the intent there really was to put memory on one (along with the bootstrap loader) and all I/O on the other. In fact, though, there was just a single Unibus with the CPU located in the middle and both ends brought out to a series of Unibus slots and then terminators. The usual PDP-11 configuration puts the CPU at one end of the Unibus with the terminator there intrinsic to the CPU implementation.

In the "upgraded" 11/45, thus in effect an 11/50 (or to a lesser degree an 11/55), the Fastbus was a dedicated interface to memory which supported two sets of memory controller plus up to four 1K or 4K (either 16- or 18-bit) bipolar semiconductor memory modules (fast 300 nS, but power hungry) - MS11-A - but later supported two sets of a different memory controller plus up to four 4Kw MOS memory modules (slower 450 nS) - MS11-B. By way of comparison a 1K bipolar module consumed 50w while a four times larger 4K MOS module consumed only 40w. These ten slots in the CPU backplane are entirely dedicated to memory; they're not Unibus (not even the two controller slots in the ten). Unibus-hosted controllers necessarily had access to this memory for data transfers but only through a special interface on the Fastbus memory controllers. See DEC-11-HMSAA-D-D for all of the details :->. Anyway, AFAIK a standard site installation would also include Unibus memory in order to populate a full address space. The Fastbus memory would exist as a well-known block within the larger memory space, where the programmer would locate key executable code in order to gain the better execution performance, whereas other code would be allocated to the slower Unibus-hosted memory.

In the 11/70 there was a unique memory subsystem that communicated with the MJ11 and/or MK11. It accessed double-words each cycle and incorporated a complex cache for additional performance improvements. Documentation simply names this the "Memory Bus". Separately there was the high-performance I/O Massbus which hosted specialized controllers for disks and tapes. And finally there was a Unibus hosting conventional controllers. AFAIK you could run an 11/70 with only Unibus-based memory but I imagine that was only an "expedient mode" operation while servicing, rather than routine operation.
 
Note that MOS Memory in later systems (e.g., the 11/34) uses the EUB (Extended Unibus) interface that is supported by the typical DD11-D 9-slot "system unit" (SU = backplane). Generally just plug-n-go, but as Michael points out you shouldn't assume that since the logical Unibus specification is agnostic in device placement (ignoring the daisy-chain interrupt handling implications) that therefore the physical implementation of any given logical Unibus is equally agnostic! Always RTFM and verify your SU. Particularly hazardous is the interplay between core/MOS memory and their backplane-connected power requirements. And then for larger memory arrays/capacities there are additional address lines that must be supplied. Caveat emptor! All of this is very much pre-PC ISA and the "modern" expectation that anywhere-goes because there is a single power specification and a single addressing specification that is always and everywhere maximally supported/implemented.

Anyway, I think that we're well past answering your question!
 
thephysicist said:
Is it possible to put a memory board into the CPU cabinet or has all memory to be in a separate cabinet?

thanks
Michael
Click to expand...
When I had my real 11/70, the MK11 memory box was in panel space 2 (counted from the top of a H960 cabinet).
Officially, it would be in a 2nd cabinet.

When the 11/70 came out, it was initially equipped with 4 BA11 boxes with core. When the MOS memory based cards came out,
the full complement of 4Mb would fit in a single BA11 box (as a single MK11)
 
You must log in or register to reply here.
Back
Top